refactor(ai): 模組化重構 - NVIDIA chat 移至 NvidiaProvider

符合 feedback_lewooogo_modular_enforcement.md 規範:
- 移除 openclaw.py 中的 _call_nvidia() (重複邏輯)
- 新增 NvidiaProvider.chat() 方法
- 更新 INvidiaProvider Protocol
- openclaw.py 改用 get_nvidia_provider().chat()
- 測試移至 test_nvidia_chat.py

架構層次:
- Router → Service → Provider (正確)
- 禁止 Service 層重複實作已存在的 Provider 功能

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

apps/api/src/services/nvidia_provider.py

@@ -92,6 +92,21 @@ class INvidiaProvider(Protocol):
         """關閉資源"""
         ...
 
+    async def chat(
+        self,
+        prompt: str,
+        model: str = ...,
+        temperature: float = ...,
+        max_tokens: int = ...,
+    ) -> tuple[str, bool, int, float]:
+        """
+        一般對話 (非 Tool Calling) - 2026-03-29 ogt 新增
+
+        Returns:
+            tuple: (response_text, success, total_tokens, cost_usd)
+        """
+        ...
+
 # =============================================================================
 # 常量定義
 # =============================================================================
@@ -635,6 +650,142 @@ class NvidiaProvider:
             if tc.valid and tc.tool_name and self.is_high_risk_tool(tc.tool_name)
         ]
 
+    async def chat(
+        self,
+        prompt: str,
+        model: str | None = None,
+        temperature: float = 0.1,
+        max_tokens: int = 2048,
+    ) -> tuple[str, bool, int, float]:
+        """
+        一般對話 (非 Tool Calling) - 用於 RCA 分析
+
+        2026-03-29 ogt: 新增，符合模組化規範
+        從 openclaw.py 遷移，統一由 NvidiaProvider 處理所有 NVIDIA API 呼叫
+
+        Args:
+            prompt: 對話內容
+            model: 模型名稱 (預設從 ModelRegistry 取得)
+            temperature: 溫度
+            max_tokens: 最大輸出 Token
+
+        Returns:
+            tuple: (response_text, success, total_tokens, cost_usd)
+        """
+        start_time = time.perf_counter()
+
+        # OTEL Span
+        with _tracer.start_as_current_span("nvidia_chat") as span:
+            span.set_attribute("ai.provider", "nvidia")
+
+            # Circuit Breaker 檢查
+            if not self._circuit_breaker.can_execute():
+                span.set_attribute("ai.error", "circuit_breaker_open")
+                NVIDIA_REQUESTS_TOTAL.labels(status="circuit_open", tool_name="chat").inc()
+                logger.warning("nvidia_chat_circuit_breaker_open")
+                return "Circuit Breaker OPEN - NVIDIA API 暫時不可用", False, 0, 0.0
+
+            # 檢查 API Key
+            if not self._api_key:
+                span.set_attribute("ai.error", "api_key_not_set")
+                return "NVIDIA_API_KEY not configured", False, 0, 0.0
+
+            # 從 ModelRegistry 取得模型
+            from src.services.model_registry import get_model_registry
+            registry = get_model_registry()
+            model_name = model or registry.get_model("nvidia", "rca")
+
+            span.set_attribute("ai.model", model_name)
+
+            logger.info(
+                "nvidia_chat_request_start",
+                model=model_name,
+                prompt_length=len(prompt),
+            )
+
+            # Langfuse 追蹤
+            with LangfuseTraceContext(
+                name="nvidia_chat",
+                metadata={"model": model_name, "task": "rca"},
+            ) as langfuse_ctx:
+                try:
+                    client = await self._get_client()
+
+                    response = await client.post(
+                        NVIDIA_API_URL,
+                        headers={
+                            "Authorization": f"Bearer {self._api_key}",
+                            "Content-Type": "application/json",
+                        },
+                        json={
+                            "model": model_name,
+                            "messages": [{"role": "user", "content": prompt}],
+                            "temperature": temperature,
+                            "max_tokens": max_tokens,
+                            "response_format": {"type": "json_object"},
+                        },
+                    )
+                    response.raise_for_status()
+                    data = response.json()
+
+                    self._circuit_breaker.record_success()
+
+                    text = data["choices"][0]["message"]["content"]
+
+                    # Token 用量
+                    usage = data.get("usage", {})
+                    prompt_tokens = usage.get("prompt_tokens", 0)
+                    completion_tokens = usage.get("completion_tokens", 0)
+                    total_tokens = usage.get("total_tokens", prompt_tokens + completion_tokens)
+
+                    # NVIDIA NIM 免費 tier = $0
+                    cost_usd = 0.0
+
+                    latency_ms = (time.perf_counter() - start_time) * 1000
+                    span.set_attribute("ai.latency_ms", latency_ms)
+                    span.set_attribute("ai.total_tokens", total_tokens)
+
+                    # Prometheus
+                    NVIDIA_REQUESTS_TOTAL.labels(status="success", tool_name="chat").inc()
+                    NVIDIA_LATENCY_SECONDS.labels(tool_name="chat").observe(latency_ms / 1000)
+
+                    # Langfuse
+                    langfuse_ctx.trace.generation(
+                        name="nvidia_chat",
+                        model=model_name,
+                        input=prompt[:500],
+                        output=text[:500],
+                        metadata={
+                            "total_tokens": total_tokens,
+                            "cost_usd": cost_usd,
+                            "latency_ms": round(latency_ms, 2),
+                        },
+                    )
+
+                    logger.info(
+                        "nvidia_chat_response_received",
+                        model=model_name,
+                        response_length=len(text),
+                        prompt_tokens=prompt_tokens,
+                        completion_tokens=completion_tokens,
+                        total_tokens=total_tokens,
+                        latency_ms=round(latency_ms, 2),
+                    )
+
+                    return text, True, total_tokens, cost_usd
+
+                except httpx.TimeoutException as e:
+                    self._circuit_breaker.record_failure()
+                    NVIDIA_REQUESTS_TOTAL.labels(status="timeout", tool_name="chat").inc()
+                    logger.warning("nvidia_chat_timeout", error=str(e))
+                    return f"Timeout: {e}", False, 0, 0.0
+
+                except Exception as e:
+                    self._circuit_breaker.record_failure()
+                    NVIDIA_REQUESTS_TOTAL.labels(status="error", tool_name="chat").inc()
+                    logger.warning("nvidia_chat_failed", error=str(e), error_type=type(e).__name__)
+                    return str(e), False, 0, 0.0
+
 
 # =============================================================================
 # 單例與工廠函數

apps/api/src/services/openclaw.py

@@ -461,76 +461,8 @@ class OpenClawService:
             logger.warning("claude_call_failed", error=str(e))
             return str(e), False
 
-    async def _call_nvidia(self, prompt: str) -> tuple[str, bool, int, float]:
-        """
-        呼叫 NVIDIA Nemotron (OpenAI 相容格式)
-
-        2026-03-29 ogt: 新增 Nemotron 一般告警支援 (非 Tool Calling)
-        2026-03-29 ogt: P1 修復 - 從 ModelRegistry 取得模型名稱
-
-        Returns:
-            tuple: (response_text, success, total_tokens, cost_usd)
-        """
-        if not settings.NVIDIA_API_KEY:
-            return "NVIDIA_API_KEY not configured", False, 0, 0.0
-
-        try:
-            client = await self._get_client()
-
-            # 從 ModelRegistry 取得模型 (P1-1 修復)
-            registry = get_model_registry()
-            model_name = registry.get_model("nvidia", "rca")
-            options = registry.get_provider_options("nvidia")
-
-            logger.info(
-                "nvidia_request_start",
-                model=model_name,
-                prompt_length=len(prompt),
-            )
-
-            response = await client.post(
-                "https://integrate.api.nvidia.com/v1/chat/completions",
-                headers={
-                    "Authorization": f"Bearer {settings.NVIDIA_API_KEY}",
-                    "Content-Type": "application/json",
-                },
-                json={
-                    "model": model_name,
-                    "messages": [{"role": "user", "content": prompt}],
-                    "temperature": options.get("temperature", 0.1),
-                    "max_tokens": options.get("max_tokens", 2048),
-                    "response_format": {"type": "json_object"},  # 強制 JSON
-                },
-                timeout=60.0,
-            )
-            response.raise_for_status()
-            data = response.json()
-
-            text = data["choices"][0]["message"]["content"]
-
-            # Token 用量
-            usage = data.get("usage", {})
-            prompt_tokens = usage.get("prompt_tokens", 0)
-            completion_tokens = usage.get("completion_tokens", 0)
-            total_tokens = usage.get("total_tokens", prompt_tokens + completion_tokens)
-
-            # NVIDIA NIM 免費 tier = $0
-            cost_usd = 0.0
-
-            logger.info(
-                "nvidia_response_received",
-                model=model_name,
-                response_length=len(text),
-                prompt_tokens=prompt_tokens,
-                completion_tokens=completion_tokens,
-                total_tokens=total_tokens,
-                cost_usd=f"${cost_usd:.6f}",
-            )
-            return text, True, total_tokens, cost_usd
-
-        except Exception as e:
-            logger.warning("nvidia_call_failed", error=str(e), error_type=type(e).__name__)
-            return str(e), False, 0, 0.0
+    # 2026-03-29 ogt: _call_nvidia 已移至 nvidia_provider.py
+    # 符合模組化規範 - 所有 NVIDIA API 呼叫統一由 NvidiaProvider 處理
 
     # =========================================================================
     # Mock LLM - Intelligent Fallback with SignOz Data
@@ -948,8 +880,10 @@ class OpenClawService:
                 elif provider == "gemini":
                     response, success, total_tokens, cost_usd = await self._call_gemini(prompt)
                 elif provider == "nvidia":
-                    # 2026-03-29 ogt: Nemotron 一般告警支援
-                    response, success, total_tokens, cost_usd = await self._call_nvidia(prompt)
+                    # 2026-03-29 ogt: 使用 NvidiaProvider.chat() (模組化規範)
+                    from src.services.nvidia_provider import get_nvidia_provider
+                    nvidia_provider = get_nvidia_provider()
+                    response, success, total_tokens, cost_usd = await nvidia_provider.chat(prompt)
                 elif provider == "claude":
                     response, success = await self._call_claude(prompt)
                 else: