docs: add gateway restart loop troubleshooting to FAQ

[otakai]

Summary

Adds a troubleshooting section to the FAQ documenting the gateway infinite restart loop caused by auto-resume of sessions that executed systemctl --user stop/restart hermes-gateway from inside the gateway process.

What is documented

• Root cause: Sessions with resume_pending=true re-execute systemctl stop hermes-gateway on every restart, creating an infinite loop.
• Fix: Clear resume_pending in sessions.json + reset systemd.
• Prevention: A ~/.local/bin/systemctl wrapper that blocks stop/restart hermes-gateway when _HERMES_GATEWAY=1 is set (the gateway sets this env var).
• Related config: agent.restart_drain_timeout recommendation.

Note that hermes gateway restart and hermes gateway stop already detect _HERMES_GATEWAY=1 and refuse to run from inside the gateway process, but systemctl --user stop/restart hermes-gateway bypasses this guard. The documented wrapper fills that gap.

[teknium1]

Thanks for documenting a real failure mode.

Automated hermes-sweeper review found that current main fixes this at the source and closes the documented bypass:

• 75ed07ace (fix(gateway): break the restart loop at the source on session resume) sanitizes an unanswered trailing tool call before replay (gateway/run.py:900), with regression coverage in tests/gateway/test_auto_continue.py:174.
• tools/terminal_tool.py:2250 now blocks gateway lifecycle commands, including raw systemctl forms, whenever _HERMES_GATEWAY=1; the wrapper proposed here is no longer needed.
• The source fix shipped in v2026.7.1.

Closing as implemented on main.