feat: enable StuckDetector for running task auto-recovery

[sh1nj1]

Problem

StuckDetector was implemented but disabled by default (enabled: false). This meant stuck tasks in running state were never auto-recovered, requiring manual intervention via Kamal console.

Additionally, pending_approval tasks had no timeout — they could remain stuck indefinitely.

Solution

1. Enable StuckDetector by default

# policy_resolver.rb
"stuck_detection" => {
  "enabled" => true,  # was: false
  ...
}

The StuckDetectorJob already runs every 10 minutes in production (configured in recurring.yml). It just wasn't doing anything because the policy was disabled.

2. Add pending_approval timeout detection

New config: approval_timeout_minutes (default: 60)

• Detects pending_approval tasks older than threshold
• Auto-recovers: marks as cancelled (vs failed for running tasks)
• Releases ResourceTracker reservation
• Dequeues next task for the topic
• Sends inbox notification with specific i18n message

Changes

• policy_resolver.rb: Enable stuck_detection, add approval_timeout_minutes
• stuck_detector.rb: Detect + auto-recover pending_approval tasks
• ai_agent.en.yml / ai_agent.ko.yml: Add approval_timeout i18n
• stuck_detector_test.rb: 3 new tests (detect, auto-recover, skip recent)

Tests

• ✅ 14 stuck_detector tests, 0 failures
• ✅ i18n completeness check passed
• ✅ Rubocop clean

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.