Monitors llama-server health with multi-phase checks (zombie detection, health endpoint, loaded model probing) and auto-restarts via systemd or manual relaunch on consecutive failures. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
