worker-reaper failing to release IN_PROGRESS tasks whose workflow completed

## Problem

worker-reaper runs every 2 minutes but is not releasing Portico tasks whose agent-dispatch workflow already completed (success OR failure). 17 tasks sat IN_PROGRESS for 12+ hours consuming all maxInFlight slots.

## Evidence

All 17 stuck tasks had matching workflow runs with `status=completed`:
- 15 with `conclusion=failure` (never released → should reap to PLANNING_VALIDATED)
- 2 with `conclusion=success` (POR-11 → PR #389 merged, POR-21 → PR #401 merged; should reap to DONE)

Dispatcher slot occupancy: 17/20 for ~12h despite queue of 205 PLANNING_VALIDATED and cron ticking every minute (`lastTickAt` current).

## Suspected causes

1. worker-reaper query does not match the workflow run to the Task (branch-name mismatch, task-id lookup, or workflow-name filter)
2. worker-reaper only reaps on timeout, not on workflow-completion; timeout threshold >12h
3. Cron may not be firing (verify with a diagnostic log line)

## Acceptance

- worker-reaper detects `completed/failure` workflow runs and reverts task to PLANNING_VALIDATED with gitBranchName=NULL within one tick (~2 min)
- worker-reaper detects `completed/success` runs and promotes task to DONE (or AWAITING_HUMAN if PR-review gate exists) within one tick
- No task remains IN_PROGRESS more than 30 min after its workflow finishes
- Emits DispatchAudit rows with outcome=`reaped_failure` / `reaped_success`

## Related

- OPER-11A (shipped) — branch-ensure unblocks fresh claims but reaper failure still stalls throughput
- Feeds into OPER-8A (dashboard) — reap counts should surface there