Add durable cleanup manifests for async task deletion

[legreffier]

Context

PR #1532 moves task deletion cleanup into async DBOS workflows. The current near-term fix can improve retries and logging around object deletion, but a full durable cleanup/reconciliation path is larger than that PR.

When task rows are deleted, task artifact and runtime session cleanup refs can disappear through FK cascade before blob/object deletion has permanently succeeded. If object cleanup later exhausts retries, operators need a durable in-app way to identify and retry orphan cleanup.

Proposal

Add a durable cleanup manifest/tombstone mechanism for task deletion and retention cleanup workflows.

Possible shape:

• Persist a cleanup manifest before destructive row deletion, including task ids, artifact object keys, runtime session object keys, requester/context, force flag, and workflow id.
• Mark manifest phases explicitly, for example: planned, rows_deleted, objects_deleted, failed, reconciled.
• Make object deletion retryable from the manifest after the original DBOS workflow has failed or aged out.
• Add a maintenance reconciler that periodically retries failed/incomplete cleanup manifests in bounded batches.
• Expose enough logs/metrics to diagnose partial cleanup without scraping DBOS internals.
• Ensure retention cleanup and explicit task deletion share the same manifest/retry model.

Acceptance criteria

• Failed object deletion after task row deletion leaves durable cleanup state with object keys.
• A maintenance workflow can retry incomplete object cleanup later.
• Cleanup work is bounded by batch size/page size, not by all artifacts/sessions attached to the deleted tasks.
• Operators can correlate cleanup attempts by manifest id/workflow id/task ids.
• Tests cover partial object deletion failure and later reconciliation.

Related

• Follow-up from async task deletion PR feat: add queue for task deletion cleanup #1532.
• Needed to fully address the durability/operability risk found in deep review around apps/rest-api/src/workflows/maintenance.ts.