Dead workers block concurrency slots indefinitely

When workers die without notifying the server, they continue holding concurrency slots. The server refuses to schedule new work until the dead worker's run is manually set to Failed. Per-deployment and global concurrency settings exacerbate this issue.

Avoid using per-deployment or global concurrency limits without runner heartbeat detection enabled. Implement timeout automations to forcibly fail stuck runs and free concurrency slots. Monitor work queue depth for blocking conditions.