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/tests/test_nvidia_chat.py

@@ -0,0 +1,159 @@
+"""
+test_nvidia_chat.py - NvidiaProvider.chat() 測試
+
+2026-03-29 ogt: 模組化重構 - 測試移至 NvidiaProvider
+符合 feedback_lewooogo_modular_enforcement.md 規範
+
+測試策略 (遵循 feedback_no_mock_testing.md):
+- 使用真實 NVIDIA API (需 NVIDIA_API_KEY)
+- 跳過條件: 無 API Key 時跳過
+"""
+
+import json
+import os
+
+import pytest
+
+from src.core.config import get_settings
+from src.services.nvidia_provider import NvidiaProvider, get_nvidia_provider
+
+settings = get_settings()
+
+
+@pytest.fixture
+def nvidia_provider():
+    """建立 NvidiaProvider 實例"""
+    return NvidiaProvider()
+
+
+@pytest.mark.asyncio
+@pytest.mark.skipif(
+    not os.getenv("NVIDIA_API_KEY") and not settings.NVIDIA_API_KEY,
+    reason="NVIDIA_API_KEY not configured",
+)
+async def test_chat_success(nvidia_provider):
+    """
+    測試 chat() 成功回應
+
+    驗證:
+    - 回應格式正確 (4-tuple)
+    - success = True
+    - total_tokens > 0
+    - cost_usd = 0 (免費 tier)
+    """
+    prompt = """你是一個 JSON 產生器。請回傳以下格式:
+    {"status": "ok", "message": "test"}
+    只回傳 JSON，不要其他內容。"""
+
+    response, success, total_tokens, cost_usd = await nvidia_provider.chat(prompt)
+
+    assert success is True, f"Expected success, got error: {response}"
+    assert isinstance(response, str)
+    assert len(response) > 0
+    assert total_tokens > 0, "Expected token count > 0"
+    assert cost_usd == 0.0, "Expected $0 for free tier"
+
+
+@pytest.mark.asyncio
+async def test_chat_no_api_key():
+    """
+    測試無 API Key 時的處理
+
+    驗證:
+    - success = False
+    - 回傳適當錯誤訊息
+    """
+    # 建立沒有 API Key 的 provider
+    provider = NvidiaProvider(api_key=None)
+
+    response, success, total_tokens, cost_usd = await provider.chat("test")
+
+    assert success is False
+    assert "not configured" in response.lower()
+    assert total_tokens == 0
+    assert cost_usd == 0.0
+
+
+@pytest.mark.asyncio
+@pytest.mark.skipif(
+    not os.getenv("NVIDIA_API_KEY") and not settings.NVIDIA_API_KEY,
+    reason="NVIDIA_API_KEY not configured",
+)
+async def test_chat_json_response(nvidia_provider):
+    """
+    測試 JSON 格式回應
+
+    驗證:
+    - 回應是有效 JSON
+    """
+    prompt = """回傳一個 JSON 物件，包含:
+    - action: "NO_ACTION"
+    - reason: "測試"
+    只回傳 JSON。"""
+
+    response, success, _, _ = await nvidia_provider.chat(prompt)
+
+    assert success is True
+
+    # 驗證是有效 JSON
+    try:
+        data = json.loads(response)
+        assert isinstance(data, dict)
+    except json.JSONDecodeError:
+        pytest.fail(f"Response is not valid JSON: {response[:200]}")
+
+
+@pytest.mark.asyncio
+@pytest.mark.skipif(
+    not os.getenv("NVIDIA_API_KEY") and not settings.NVIDIA_API_KEY,
+    reason="NVIDIA_API_KEY not configured",
+)
+async def test_chat_uses_model_registry(nvidia_provider):
+    """
+    測試使用 ModelRegistry 取得模型
+
+    驗證:
+    - 使用 models.json 中定義的 rca 模型
+    """
+    from src.services.model_registry import get_model_registry
+
+    registry = get_model_registry()
+    expected_model = registry.get_model("nvidia", "rca")
+
+    # 模型應該是 llama-3.1-nemotron-70b-instruct
+    assert "nemotron" in expected_model.lower()
+    assert "70b" in expected_model or "mini" in expected_model
+
+
+@pytest.mark.asyncio
+def test_get_nvidia_provider_singleton():
+    """
+    測試單例模式
+
+    驗證:
+    - get_nvidia_provider() 返回同一實例
+    """
+    provider1 = get_nvidia_provider()
+    provider2 = get_nvidia_provider()
+
+    assert provider1 is provider2
+
+
+@pytest.mark.asyncio
+@pytest.mark.skipif(
+    not os.getenv("NVIDIA_API_KEY") and not settings.NVIDIA_API_KEY,
+    reason="NVIDIA_API_KEY not configured",
+)
+async def test_chat_includes_otel_tracing(nvidia_provider):
+    """
+    測試 OTEL 追蹤整合
+
+    驗證:
+    - chat() 執行時有 OTEL span
+    """
+    # 這個測試主要驗證代碼不會拋出異常
+    prompt = '{"test": true}'
+    response, success, _, _ = await nvidia_provider.chat(prompt)
+
+    # 只要沒有拋出異常就算通過
+    assert isinstance(response, str)