fix: Bug A 診斷 + Bug B 真修 — LLM 120s/130s 硬編 → OPENCLAW_TIMEOUT

殘留兩個深層 bug 處理:

Bug A (approval.incident_id 仍 NULL) — 加診斷
  - update_incident_id 加 rowcount 檢查
  - 若 UPDATE 0 rows affected → warning log (id 型別 mismatch 或 session 不同步)
  - 手動 UPDATE 測試通過 → DB/permissions 正常，問題在應用層
  - 等 CD 部署後 live-fire 觀察 log 診斷真因

Bug B (LLM 仍 2m6s >> 30s) — 真修
  openclaw.py 兩處硬編 timeout:
  - line 146 httpx client default: 120.0s → settings.OPENCLAW_TIMEOUT (30s)
  - line 348 /analyze/incident POST: 130.0s → settings.OPENCLAW_TIMEOUT (30s)
  GAP-B4 commit dd0a778 只修了 ai_providers/ollama.py
  但 openclaw.py 自己的 httpx client 和 endpoint call 沒改
  這就是為什麼 Live-fire #2-#7 都卡 120s+ 的真因

回歸測試: 125/125 (dispatcher + a4 + classify + grouping)

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

apps/api/src/services/approval_db.py

@@ -606,13 +606,33 @@ class ApprovalDBService:
         """
         2026-04-06 ogt: Phase 26 — 回寫 incident_id 到 approval_records
         讓 Playbook 萃取和 KM 寫入能找到對應的 Incident
+
+        2026-04-14 Claude Sonnet 4.6 診斷: Live-fire #7 發現 approval.incident_id 仍 NULL
+        加 rowcount 與 pre/post 值檢查，若 0 rows affected 則 log warning
         """
         async with get_db_context() as db:
-            await db.execute(
+            result = await db.execute(
                 update(ApprovalRecord)
                 .where(ApprovalRecord.id == str(approval_id))
                 .values(incident_id=incident_id)
             )
+            rowcount = result.rowcount if hasattr(result, "rowcount") else -1
+            if rowcount == 0:
+                # 找不到對應 approval — 可能 id 型別或 session 不同步
+                logger.warning(
+                    "update_incident_id_zero_rows",
+                    approval_id=str(approval_id),
+                    approval_id_type=type(approval_id).__name__,
+                    incident_id=incident_id,
+                    reason="UPDATE 0 rows affected — approval 不存在或 id mismatch",
+                )
+            else:
+                logger.info(
+                    "update_incident_id_success",
+                    approval_id=str(approval_id),
+                    incident_id=incident_id,
+                    rowcount=rowcount,
+                )
 
     async def update_telegram_message(
         self, incident_id: str, telegram_message_id: int, telegram_chat_id: int | None = None

apps/api/src/services/openclaw.py

@@ -140,10 +140,15 @@ class OpenClawService:
         self._signoz = get_signoz_client()
 
     async def _get_client(self) -> httpx.AsyncClient:
-        """取得 HTTP 客戶端"""
+        """取得 HTTP 客戶端
+
+        2026-04-14 Claude Sonnet 4.6: 從硬編 120s 改用 OPENCLAW_TIMEOUT 設定 (30s)
+        對齊 ADR-052 GAP-B4 的 25s + 5s buffer 設計。原 120s 違反 defense-in-depth。
+        """
         if self._http_client is None or self._http_client.is_closed:
+            _t = float(settings.OPENCLAW_TIMEOUT)
             self._http_client = httpx.AsyncClient(
-                timeout=httpx.Timeout(120.0, connect=10.0),
+                timeout=httpx.Timeout(_t, connect=10.0),
             )
         return self._http_client
 
@@ -342,10 +347,12 @@ class OpenClawService:
                 "affected_services": affected_services,
                 "expert_context": _to_serializable(expert_context) if expert_context else None,
             }
+            # 2026-04-14 Claude Sonnet 4.6: 從硬編 130s 改用 OPENCLAW_TIMEOUT
+            # 原 130s 讓 LLM 能卡 2m10s，超過 Ollama 真實返回時間（P95 54s）
             resp = await client.post(
                 f"{settings.OPENCLAW_URL}/api/v1/analyze/incident",
                 json=payload,
-                timeout=httpx.Timeout(130.0, connect=5.0),
+                timeout=httpx.Timeout(float(settings.OPENCLAW_TIMEOUT), connect=5.0),
             )
             resp.raise_for_status()
             data = resp.json()