[Refactor] Give Fast parent events one delivery lifecycle

[roomote-roomote]

​Created by Roomote. View the task or mention @roomote-roomote for follow-up asks.

What changed

• Route Fast parent pull-request, settlement, and artifact events through one leased claim, delivery, and finalization lifecycle.
• Fence lifecycle writes with unique claim tokens so an expired claimant cannot overwrite or release a newer lease.
• Move Fast failed-start retry ancestry, budget, backoff, idempotency, relaunch, and lifecycle recording into the task-run queue operation.
• Preserve duplicate suppression and permanent, transient, and post-Slack delivery classifications with focused race coverage.

Why this change was made

Fast parent event adapters had independently accumulated the same stateful delivery lifecycle, while settlement notification also owned task relaunch policy. Keeping those responsibilities in multiple presentation paths made failure handling and retry behavior likely to drift.

Impact

There is no intended user-facing behavior change. Fast child events now have one lifecycle owner, and concurrent or stale delivery and retry attempts resolve consistently without duplicating parent updates or child relaunches.

[roomote-community]

No code issues found. See task

_{Reviewed cc0f9ad}