Workers stop when relaunch.sh rebuilds the app mid-orchestration

[PureWeen]

Problem

When a multi-agent worker runs relaunch.sh (or the app is rebuilt/relaunched for any reason), the orchestration loop loses track of the worker. The worker session gets a new SDK session after relaunch, but the orchestrator's SendPromptAndWaitAsync was awaiting the old TCS which gets canceled/orphaned.

Observed behavior

In the PP- IC Things orchestration:

• Worker-1 was dispatched and started doing work
• Worker called relaunch.sh to rebuild PolyPilot with code changes
• After relaunch, the worker's session was restored but the orchestrator never received the completion signal
• The orchestrator's reflection loop hung waiting for a worker result that would never come

Expected behavior

When the app relaunches mid-orchestration:

1. The pending orchestration should be detected and resumed via PendingOrchestration
2. Worker results from before the relaunch should be recoverable
3. OR the orchestrator should detect the relaunch and re-dispatch the worker

Notes

This is related to but distinct from the server idle timeout issue (#396). In this case the app itself is restarted, not just the server killing an idle session.