fix(ai): isolate heavy Ollama workloads from GCP alert lane

apps/api/src/services/log_summary_service.py

@@ -35,8 +35,9 @@ logger = structlog.get_logger(__name__)
 # ============================================================
 # 2026-05-03 ogt: ADR-110 GCP-A Primary — 改從 settings 讀取，不再硬編碼 111
 def _get_ollama_url() -> str:
-    from src.core.config import get_settings
-    return get_settings().OLLAMA_URL
+    from src.services.ollama_endpoint_resolver import resolve_ollama_endpoint
+
+    return resolve_ollama_endpoint("deep_rca")
 # D1 集中化 2026-04-11: 從 models.json providers.ollama.models.log_anomaly 讀取
 SUMMARY_MODEL = get_model("ollama", "log_anomaly")
 LLM_TIMEOUT = 180.0       # deepseek-r1 硬超時
@@ -145,7 +146,7 @@ class LogSummaryService:
                 self.summarize(pod_name, namespace),
                 timeout=SOFT_TIMEOUT,
             )
-        except asyncio.TimeoutError:
+        except TimeoutError:
             logger.info(
                 "log_summary_soft_timeout",
                 pod=pod_name,
@@ -182,7 +183,7 @@ class LogSummaryService:
     def _extract_anomaly_lines(self, raw_logs: str) -> list[str]:
         """過濾異常行 + 敏感資料遮蔽，取最後 ANOMALY_TAIL_LINES 行"""
         lines = raw_logs.splitlines()
-        anomaly = [l for l in lines if _ANOMALY_PATTERN.search(l)]
+        anomaly = [line for line in lines if _ANOMALY_PATTERN.search(line)]
         # 取最後 N 行
         anomaly = anomaly[-ANOMALY_TAIL_LINES:]
         # 遮蔽敏感資料