[EPIC] Implement retention-driven purge for IRC tables

### Describe the proposal

Implement the accepted handoff from an expired retained deletion to asynchronous physical Iceberg purge. The implementation must safely select eligible records, claim work, preserve the lifecycle boundary during worker retries, and reuse existing hard-delete mechanics where the approved design permits.

### Task list

- [ ] Implement expiry selection and purge-work claiming.
- [ ] Integrate the approved handoff with cleanup execution.
- [ ] Implement restart, retry, and multi-replica safety behavior.
- [ ] Add operational logging, metrics, and guidance.
- [ ] Add end-to-end expiry and purge-race coverage.

Depends on: #12393, #12394, #12395

Related to: #12392, `design-docs/async-iceberg-rest-hard-deletion.md`