Fixes #60042.

ShouldBeUniqueUntilProcessing jobs need to handle two different retry scenarios:

• If the unique lock was released before handling and the job later fails, a retry must not release a lock acquired by a newer dispatch.
• If middleware releases the job before handling, the unique lock remains held and must be released when a later attempt is eventually processed.

Using attempts() <= 1 cannot distinguish these scenarios. It prevents retries from releasing newer locks, but also causes middleware-released jobs to retain their original lock indefinitely.

This change tracks the actual unique lock owner on Queueable jobs and uses ownership-aware lock release:

• Records the cache lock owner when the unique lock is acquired.
• Serializes the owner with the queued job so it survives retries.
• Releases the lock through restoreLock(...)->release().
• Allows later attempts to release a lock when they carry its owner.
• Retains the first-attempt guard for legacy and ownerless payloads.
• Propagates the owner through hidden job context so missing-model cleanup cannot release a replacement lock.
• Supports queued listeners and custom uniqueVia() stores.