Problem

When a gateway session gets stuck (hung terminal command, runaway tool loop), restarting the gateway re-enters the stuck state. The user cannot get a clean start.

What happens:

1. Agent enters a stuck state (hung terminal command, tool loop, 30+ minute hang)
2. User restarts gateway (hermes gateway stop && hermes gateway)
3. Telegram redelivers the last unacknowledged message
4. Agent loads the session from the DB, sees the conversation context, and resumes the stuck task
5. Back to step 1

User report: "It keeps trying to resume the previous task every time I reboot the gateway and it ends up dying. It's unusable."

The user's only escape was /stop (force-stop) which eventually worked, but required the agent to be responsive enough to process the command.

Current workaround

hermes gateway stop
hermes sessions delete <stuck-session-id>
hermes gateway

Or nuclear: rm -f ~/.hermes/hermes_state.db (regenerates on start, loses all session history).

Suggested fixes

1. /stop should mark the session as "don't resume" — set a flag in the session DB so the gateway skips it on restart
2. hermes gateway --clean — start with no pending sessions, ignore Telegram's update backlog
3. Detect stuck resume loops — if the same session enters the error/timeout state within N seconds of gateway start, abandon it instead of retrying
4. Consume stale Telegram updates on startup — on gateway start, read and discard any Telegram updates older than N minutes