What does this PR do?

A worker can produce some requested outputs, record their PR URLs, and then time out with work still unfinished. The dispatcher returns the card to ready , but active_pr suppresses every retry for 24 hours. An unrelated PR reference can cause the same hold.

Remove that comment-regex guard and its handoff exceptions. Dispatch continues to use the existing task lifecycle, dependency checks, atomic claims, worker ownership, failure limits, authentication guard, recent-success guard, and cooldowns.

Retry context now gives general recovery guidance: inspect and verify existing work against the requested outcome, reuse completed work, continue what remains, and record completion when the full outcome is satisfied. This is model guidance, not a guarantee of exactly-once external side effects.…