feat(p3.1-t2-patha): DiagnosisAggregator 路徑 A + Solver F4 critical reject + 對齊測試

Wave 8 P3.1-T2 PathA 啟用 + Solver F4 安全強化 + test 對齊：

PathA — DiagnosisAggregator 信號分類層補 PDI:
- ENABLE_DIAGNOSIS_AGGREGATOR default=False → True
  · PathA 純信號分類層（OOMKilled/CrashLoop 等業務邏輯）
  · 不重複呼叫 K8s/SignOz API（只取 PDI 已收集的 raw 資料）
  · 安全 default on — 純邏輯處理，無外部依賴重疊
- diagnosis_aggregator.py +155 行（PathA 實作）
- pre_decision_investigator.py 已接 (commit 3a2cd151)

F4 — Solver critical risk reject:
- solver_agent.py: _validate_recommended_action 拒絕 risk=critical
  · 鐵律：critical 動作必須走人工審批，不可變 Telegram 按鈕
  · log warning + return None（被 _extract 過濾掉）
- _extract_recommended_actions 改返回 (list, status_str) tuple
  · status="ok"/"empty"/"all_invalid" 供呼叫端決策
- protocol.py +16 / metrics.py +9 / ai_router.py +18 — 配套 metric + protocol field

測試對齊:
- test_solver_recommended_actions.py 拆 test_all_valid → low/medium/high accepted +
  test_critical_rejected
- result tuple unpack: result, _ = _extract_recommended_actions(...)
- test_diagnosis_aggregator_stub.py: feature flag default 改 True 對齊 PathA

Tests: 51 passed (solver 28 + aggregator 16 + router fallback 8)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-Authored-By: Multiple Engineers (Wave 8 P3.1-T2 PathA + F4) <noreply@anthropic.com>

apps/api/src/services/ai_router.py

@@ -1094,6 +1094,10 @@ class AIRouterExecutor:
         for provider_name in provider_order:
             # 2026-04-27 Claude Sonnet 4.6: A2 — 若上一輪失敗且本輪開始，表示發生 fallback
             # 記錄 metric（DIAGNOSE intent 專屬；非 DIAGNOSE 不記，不影響其他路徑）
+            # 2026-04-27 Claude Sonnet 4.6: F6 — fallback metric 只在真實 analyze() 失敗時觸發
+            # _last_attempted_provider 僅在 provider.analyze() 執行後失敗才賦值（見下方兩處）；
+            # not_registered / privacy_skip / circuit_open / rate_limit 分支均不賦值，
+            # 避免這些「被跳過的 provider」誤計入 from→to fallback 鏈，metric 不可信問題（F6）。
             if _is_diagnose_intent and _last_attempted_provider is not None:
                 try:
                     from src.core.metrics import record_diagnose_fallback
@@ -1107,16 +1111,24 @@ class AIRouterExecutor:
                         to_provider=provider_name,
                     )
                 except Exception as _metric_e:
-                    logger.debug("diagnose_fallback_metric_failed", error=str(_metric_e))
+                    # 2026-04-27 Claude Sonnet 4.6: F6 — 升 warning（原 debug 會 silent swallow）
+                    # + 計入 error counter 讓 metric 管道問題可被 Prometheus 偵測
+                    logger.warning("diagnose_fallback_metric_failed", error=str(_metric_e))
+                    try:
+                        from src.core.metrics import AIOPS_DIAGNOSE_FALLBACK_METRIC_ERROR_TOTAL
+                        AIOPS_DIAGNOSE_FALLBACK_METRIC_ERROR_TOTAL.inc()
+                    except Exception:
+                        pass
 
             provider = self._registry.get(provider_name)
             if not provider:
                 # 2026-04-14 Claude Sonnet 4.6: silent skip 改 errors 累積（觀測性）
+                # 2026-04-27 Claude Sonnet 4.6: F6 — 不設 _last_attempted_provider（未真實執行 analyze）
                 errors.append(f"{provider_name}: not_registered")
-                _last_attempted_provider = provider_name
                 continue
 
             # 隱私過濾 (D7)
+            # 2026-04-27 Claude Sonnet 4.6: F6 — privacy_skip 不設 _last_attempted_provider（未嘗試）
             if require_local and provider.privacy_level != "local":
                 errors.append(f"{provider_name}: privacy_skip(non_local)")
                 continue
@@ -1126,7 +1138,7 @@ class AIRouterExecutor:
             if cb.is_open():
                 errors.append(f"{provider_name}: circuit_open")
                 logger.warning("ai_router_circuit_open", provider=provider_name)
-                _last_attempted_provider = provider_name
+                # 2026-04-27 Claude Sonnet 4.6: F6 — circuit_open 不設 _last_attempted_provider（未嘗試）
                 continue
 
             # 閘門 2: Rate Limiter

apps/api/src/services/diagnosis_aggregator.py

@@ -556,6 +556,161 @@ class DiagnosisAggregator:
                 },
             ))
 
+    def classify_signals_from_raw(
+        self,
+        k8s_data: dict | None = None,
+        logs_data: str | None = None,
+        metrics_data: dict | None = None,
+    ) -> list[DiagnosisSignal]:
+        """
+        2026-04-27 P3.1-T2-PathA by Claude — DiagAggregator 信號分類層補 PDI
+
+        純邏輯信號分類：接受 PDI 已收集的 raw 資料做業務邏輯分類，
+        不打外部 API（K8s/SignOz），不重複收集。
+
+        Args:
+            k8s_data:    EvidenceSnapshot.k8s_state（D1，PDI 已收集的 dict）
+            logs_data:   EvidenceSnapshot.recent_logs（D2，sanitized string）
+            metrics_data: EvidenceSnapshot.metrics_snapshot（D3，PDI 已收集的 dict）
+
+        Returns:
+            list[DiagnosisSignal]: 分類後的信號清單（空清單代表無異常）
+        """
+        # 組裝暫時 context 供 _analyze_signals 使用（不觸發任何 IO）
+        ctx = DiagnosisContext(target="_classify_only")
+
+        # D1: k8s_state dict → 嘗試映射為 K8sDiagnostics（只提取可分類欄位）
+        if k8s_data and isinstance(k8s_data, dict):
+            # 利用 K8sDiagnostics.from_dict（若存在）或直接從常見欄位提取信號
+            # 不依賴 K8sDiagnostics.from_dict（避免 import coupling），
+            # 改從 k8s_data 中提取已知信號模式
+            self._classify_k8s_dict_signals(ctx, k8s_data)
+
+        # D3: metrics_snapshot → GoldMetrics-like 分析
+        if metrics_data and isinstance(metrics_data, dict):
+            self._classify_metrics_dict_signals(ctx, metrics_data)
+
+        # D2: logs string → 錯誤計數分類
+        if logs_data and isinstance(logs_data, str):
+            self._classify_log_string_signals(ctx, logs_data)
+
+        return ctx.signals
+
+    def _classify_k8s_dict_signals(
+        self,
+        context: DiagnosisContext,
+        k8s_data: dict,
+    ) -> None:
+        """
+        2026-04-27 P3.1-T2-PathA by Claude — 從 PDI k8s_state dict 提取信號
+        不依賴 K8sDiagnostics 物件，直接從 dict 關鍵字段分類。
+        """
+        phase = str(k8s_data.get("phase", "")).lower()
+        reason = str(k8s_data.get("reason", "")).lower()
+        restart_count = k8s_data.get("restart_count", 0) or 0
+
+        # CrashLoopBackOff
+        if "crashloop" in phase or "crashloopbackoff" in reason:
+            context.signals.append(DiagnosisSignal(
+                source="k8s_state",
+                signal_type="crash_loop",
+                severity=DiagnosisSeverity.CRITICAL,
+                message=f"CrashLoopBackOff detected (phase={k8s_data.get('phase', '?')})",
+                evidence={"phase": k8s_data.get("phase"), "reason": k8s_data.get("reason")},
+            ))
+
+        # OOMKilled
+        if "oomkilled" in phase or "oomkilled" in reason or "oom" in reason:
+            context.signals.append(DiagnosisSignal(
+                source="k8s_state",
+                signal_type="oom_killed",
+                severity=DiagnosisSeverity.CRITICAL,
+                message=f"OOMKilled detected (reason={k8s_data.get('reason', '?')})",
+                evidence={"phase": k8s_data.get("phase"), "reason": k8s_data.get("reason")},
+            ))
+
+        # Image pull error
+        if "imagepullerr" in reason or "errimagepull" in reason or "imagepullbackoff" in reason:
+            context.signals.append(DiagnosisSignal(
+                source="k8s_state",
+                signal_type="image_pull_error",
+                severity=DiagnosisSeverity.HIGH,
+                message=f"Image pull error (reason={k8s_data.get('reason', '?')})",
+                evidence={"reason": k8s_data.get("reason")},
+            ))
+
+        # High restart count
+        try:
+            rc = int(restart_count)
+        except (TypeError, ValueError):
+            rc = 0
+        if rc > 5:
+            context.signals.append(DiagnosisSignal(
+                source="k8s_state",
+                signal_type="high_restart_count",
+                severity=DiagnosisSeverity.MEDIUM,
+                message=f"High restart count: {rc}",
+                evidence={"restart_count": rc},
+            ))
+
+    def _classify_metrics_dict_signals(
+        self,
+        context: DiagnosisContext,
+        metrics_data: dict,
+    ) -> None:
+        """
+        2026-04-27 P3.1-T2-PathA by Claude — 從 PDI metrics_snapshot dict 提取信號
+        """
+        try:
+            error_rate = float(metrics_data.get("error_rate", 0) or 0)
+        except (TypeError, ValueError):
+            error_rate = 0.0
+
+        try:
+            p99_ms = float(metrics_data.get("p99_latency_ms", 0) or 0)
+        except (TypeError, ValueError):
+            p99_ms = 0.0
+
+        if error_rate > 5:
+            context.signals.append(DiagnosisSignal(
+                source="metrics_snapshot",
+                signal_type="high_error_rate",
+                severity=DiagnosisSeverity.CRITICAL if error_rate > 20 else DiagnosisSeverity.HIGH,
+                message=f"High error rate: {error_rate:.2f}%",
+                evidence={"error_rate": error_rate},
+            ))
+
+        if p99_ms > 5000:
+            context.signals.append(DiagnosisSignal(
+                source="metrics_snapshot",
+                signal_type="high_latency",
+                severity=DiagnosisSeverity.HIGH if p99_ms >= 10000 else DiagnosisSeverity.MEDIUM,
+                message=f"High P99 latency: {p99_ms:.0f}ms",
+                evidence={"p99_latency_ms": p99_ms},
+            ))
+
+    def _classify_log_string_signals(
+        self,
+        context: DiagnosisContext,
+        logs_data: str,
+    ) -> None:
+        """
+        2026-04-27 P3.1-T2-PathA by Claude — 從 PDI recent_logs string 提取信號
+        """
+        # 簡單計數 ERROR/FATAL 行
+        error_lines = [
+            line for line in logs_data.splitlines()
+            if any(kw in line.upper() for kw in ("ERROR", "FATAL", "CRITICAL", "EXCEPTION", "TRACEBACK"))
+        ]
+        if len(error_lines) > 10:
+            context.signals.append(DiagnosisSignal(
+                source="recent_logs",
+                signal_type="frequent_errors",
+                severity=DiagnosisSeverity.HIGH if len(error_lines) >= 50 else DiagnosisSeverity.MEDIUM,
+                message=f"Frequent error lines in logs: {len(error_lines)}",
+                evidence={"error_line_count": len(error_lines), "sample": error_lines[:3]},
+            ))
+
     # =========================================================================
     # Utilities
     # =========================================================================