Async actor result can be lost after parent re-entry fails

A finished spawned actor is marked as reintegrated before the parent synthetic turn starts. If parent re-entry rejects, the queue swallows the error. Every later drain skips the result, so it is lost for the rest of that service-worker lifetime.

A simple flag rollback is not sufficient. Re-entry can reject after the synthetic message reaches the transcript. A retry can then append the same actor reply twice.

Required behavior:

• Keep one stable delivery identity for the actor-result batch.
• Claim one delivery attempt at a time.
• Mark delivery complete only after durable transcript acceptance.
• Retry with the same identity after an ambiguous failure.
• Deduplicate the transcript append by that identity.
• Recheck the parent Stop generation before each retry.

Start by locating spawned-actor reintegration, the parent synthetic turn, queue draining, transcript acceptance, and Stop-generation checks. Trace how the actor-result batch is marked complete and retried; done means one stable identity is claimed per attempt, durable acceptance completes delivery, ambiguous failures retry without duplicate transcript entries, and Stop generation is rechecked.