fix(taskctl): jobs/tasks go stale after binary restart — need auto-recovery on startup

[randomm]

Problem

When opencode is restarted (new binary), any running taskctl jobs are left in a stale state:

• Job remains running with a dead Pulse PID
• Tasks stuck in developing/reviewing/adversarial-running with no active agents
• taskctl resume refuses because job is still marked running
• taskctl stop sends signal but Pulse is dead so nothing processes it
• enableAutoWakeup() in-memory listener dies with the process — PM never wakes up even after restart
• User must manually edit JSON files to force job to stopped

Fix

On taskctl start and taskctl resume, run a recovery scan that:

1. Checks if the job's Pulse PID is alive — if not, immediately marks job stopped
2. Resets tasks in developing/reviewing/adversarial-running back to open
3. Cleans up orphaned worktrees for those tasks
4. Re-calls enableAutoWakeup(pmSessionId) so PM wakeup listener is re-registered
5. Logs a system comment on each reset task

Also: taskctl stop on a dead-PID job should immediately mark stopped rather than sending signal.

Acceptance Criteria

• After binary restart, taskctl resume <jobId> auto-detects dead PID, resets stale tasks, restarts Pulse
• taskctl stop on dead-PID job immediately transitions to stopped
• Tasks in flight at restart are reset to open with a system comment
• enableAutoWakeup re-registered on resume so PM wakes up on task completion
• No manual JSON editing needed after restart
• Tests cover restart/recovery scenario
• Adversarial: APPROVED