Pipeline scheduler: recover stale InProgress tasks after migration

Problem

After moving from the old dev/verify/integrate loop to the scheduler, stale InProgress tasks from the legacy pipeline can remain stranded indefinitely.

Observed example:

• t-647 is still InProgress with an old pipeline verification-failure comment
• scheduler only consumes task ready + Open candidates, so these tasks are never re-queued automatically

Impact

• Work can be permanently stuck without manual intervention
• Queue health degrades over time as legacy state accumulates

Proposed fix

• Add startup reconciliation for stale task statuses:
• detect InProgress tasks with no active scheduler run and old updated_at
• transition back to Open (or NeedsHelp) with system comment
• optionally reconcile stale pipeline_runs rows at the same time

Acceptance criteria

• Stale InProgress tasks are automatically recovered to actionable states
• Scheduler does not leave legacy pipeline artifacts blocking task flow