fix(taskctl): use assignee_pid for liveness in resurrectionScan and heartbeatActiveAgents

[randomm]

Problem

Both resurrectionScan and heartbeatActiveAgents in pulse.ts use isSessionActivelyRunning(task.assignee) to check if a developer session is alive. This reads from SessionStatus, a module-level in-memory Map.

After a binary restart, SessionStatus can contain stale busy entries for sessions from before the restart. This causes tasks to get stuck at stage: developing forever after restart.

Root Cause

SessionStatus state is not cleared on restart. isPidAlive(task.assignee_pid) is the authoritative signal.

Fix

Add PID-based liveness check in both resurrectionScan and heartbeatActiveAgents:

• Dead if EITHER SessionStatus is idle OR assignee PID is dead
• In resurrectionScan: if stage === developing and dead, advance to reviewing (preserve worktree) instead of resetting to idle

Acceptance Criteria

• After binary restart, task with stage: developing and dead assignee PID advances to reviewing within one Pulse tick
• Worktree preserved when advancing to reviewing
• bun test and bun run typecheck pass

Closes #252, #253

[randomm]

Resolved by PR #255 (merged ec304a5).