fix(worker): 防止事务泄漏并恢复滞留队列

为 PostgreSQL 连接增加可配置的事务空闲与锁等待超时，并在请求取消或 Worker 退出后使用独立有界上下文回滚事务。\n\n恢复过期任务时按批次使用 SKIP LOCKED，周期重建缺失或终态 River Job；锁等待超时只在确认尚未提交上游时安全重排，避免重复执行和重复结算。\n\n验证：完整 Go 测试、go vet、生产 Kustomize 渲染、gofmt 与 git diff --check 均通过。

apps/api/internal/store/worker_registry.go

@@ -70,7 +70,7 @@ func (s *Store) RegisterWorkerInstance(ctx context.Context, input WorkerRegistra
 	if err != nil {
 		return WorkerAllocation{}, err
 	}
-	defer tx.Rollback(ctx)
+	defer rollbackTransaction(tx)
 	// Worker heartbeats and allocations are ephemeral coordination state that is
 	// refreshed every few seconds. Do not hold the global allocation lock while
 	// waiting for a synchronous replica to acknowledge the commit.
@@ -245,7 +245,7 @@ func (s *Store) RecoverOrphanedAsyncRiverJobs(
 	if err != nil {
 		return 0, err
 	}
-	defer tx.Rollback(ctx)
+	defer rollbackTransaction(tx)
 	if _, err := tx.Exec(ctx, `SELECT pg_advisory_xact_lock(hashtextextended('gateway-river-orphan-recovery', 0))`); err != nil {
 		return 0, err
 	}