Startup recovery leaves queued jobs leased when re-enqueue fails

Bug Description

During BackgroundExecutionService.sweep_orphans_on_startup(), each queued workflow is first claimed with claim_queued_lease(). If request reconstruction succeeds but _enqueue() raises, the exception is silently suppressed and the lease is not released.

The row remains QUEUED with a fresh owner/heartbeat even though it was never placed in the executor. In the default in-process backend, the orphan watchdog is not started, and the startup sweep is one-shot. The job can therefore remain invisible to recovery until another process restart (and until its lease becomes stale).

Reproduction

1. Return one queued workflow from _queued_workflow_jobs().
2. Make claim_queued_lease() succeed.
3. Make _reconstruct_request() succeed.
4. Fault-inject _enqueue() to raise.
5. Run sweep_orphans_on_startup().
6. Observe that release_queued_lease() was never called and the exception was swallowed.

Expected behavior

If enqueue fails after a queued lease was claimed, log the failure and release that exact lease using its owner and heartbeat token. This restores the job's recoverability without allowing a stale claimant to clear a newer lease.

Environment

• Branch: release-1.13.0
• Commit checked: 05b6498e8b54c2d89042afc2e2a96c452dd12002
• Backend: default in-process execution
• Python: 3.12