[prathshenoy]

Summary

What:

• Resume reconciliation through one bounded batch per delivery, acknowledge completed owning notifications, and postpone unfinished or borrowed work.
• Reject superseded sources and leave runtime consumers unregistered until old workers cannot write after takeover and terminal recovery is implemented.

Why:

• Continue long scans without restarting them or repeatedly starting already-completed refreshes.
• Keep message coordination separate from issue mutations and verify the redelivery path against a real queue.

Test Plan

• Add unit tests.
• Verify chunk continuation, completion ownership, and late refresh handling with a MySQL-backed queue and storage.

Revert Plan

• Revert dependents first, then this PR. No runtime consumer is enabled; the bounded reconciliation and storage layers remain unused by the reference host.

Issue Links

• CODEM-607

Stack

1. feat(stovepipe): Store issue-tracking progress #930
2. feat(stovepipe): Reconcile issues in batches #932
3. feat(stovepipe): Orchestrate issue checkpoints #940
4. feat(stovepipe): Publish issue-tracking updates #939