fix(cluster): 修复身份事务阻塞与 Worker 容量抖动

将身份协调和安全事件心跳改为 PostgreSQL 单 Leader 执行，并从独立 Worker 进程中移除身份运行时，避免多副本重复写同一状态。

为安全事件事务增加锁等待、空闲事务超时及独立回滚上下文；Worker 需连续丢失六次心跳后才判定失效，降低跨节点抖动导致的容量反复扩缩。

验证：Go 全量测试、go vet、Race 聚焦测试、PostgreSQL 18 Leader/安全事件/Worker 分配集成测试及迁移测试通过。

apps/api/internal/runner/queue_worker.go

@@ -215,12 +215,13 @@ func (s *Service) loadAsyncWorkerCapacity(ctx context.Context) (store.AsyncWorke
 		return store.AsyncWorkerCapacitySnapshot{}, err
 	}
 	allocation, err := s.store.RegisterWorkerInstance(ctx, store.WorkerRegistrationInput{
-		InstanceID:      s.workerInstanceID,
-		PodUID:          strings.TrimSpace(os.Getenv("POD_UID")),
-		PodName:         strings.TrimSpace(os.Getenv("POD_NAME")),
-		Site:            strings.TrimSpace(os.Getenv("EASYAI_SITE")),
-		Revision:        strings.TrimSpace(os.Getenv("AI_GATEWAY_REVISION")),
-		DesiredCapacity: snapshot.Capacity,
+		InstanceID:          s.workerInstanceID,
+		PodUID:              strings.TrimSpace(os.Getenv("POD_UID")),
+		PodName:             strings.TrimSpace(os.Getenv("POD_NAME")),
+		Site:                strings.TrimSpace(os.Getenv("EASYAI_SITE")),
+		Revision:            strings.TrimSpace(os.Getenv("AI_GATEWAY_REVISION")),
+		DesiredCapacity:     snapshot.Capacity,
+		HeartbeatStaleAfter: time.Duration(s.cfg.AsyncWorkerRefreshIntervalSeconds) * 6 * time.Second,
 	})
 	if err != nil {
 		return store.AsyncWorkerCapacitySnapshot{}, err