Feature: Graceful shutdown

[SeanathanVT]

Summary

Comprehensive rewrite of the shutdown path to ensure the belt always stops and BLE disconnects cleanly, plus a web UI notification so users see "Server is shutting down" instead of their browser silently going dead.

Problem

• Ctrl+C on run.py sent SIGINT to both the wrapper and Waitress subprocess (same process group), causing race conditions: connection reset errors, belt potentially left running, no UI notification
• Clicking Close while BLE scanning was still in progress crashed with RuntimeError: Event loop stopped before Future completed
• _deferred_exit only waited 1s, not enough time for the browser's /stats poll to arrive and receive the shutdown signal

Changes

app.py

• Added _server_stopping flag exposed via /stats endpoint so all session templates can detect shutdown
• Increased deferred exit delay from 1s to 5s, giving browser time (worst case 3s poll interval) to receive /stats with stopping: true
• Wrapped loop.run_until_complete(_connect_to_pad()) in try/except for RuntimeError to catch the race when shutdown interrupts BLE scanning

run.py

• Added preexec_fn=os.setsid() to subprocess.Popen() so Waitress runs in its own process group. Ctrl+C only hits run.py
• POST /shutdown before terminating, with 15s timeout (BLE cleanup can take a few seconds)
• Added 6s wait after shutdown call so server exits cleanly on its own without forced terminate

Templates

active_session.html, paused_session.html, start_session.html all check data.stopping === true in their /stats poll callback and display "Server is shutting down. You may close this window."

Documentation

• CHANGELOG.md: added shutdown notification and BLE scanning race condition fixes
• ROADMAP.md: updated section 1.2 with accurate implementation details (os._exit, UI notification, process isolation)
• README.md: updated graceful shutdown description

Shutdown Flow (Ctrl+C)

1. run.py catches KeyboardInterrupt
2. POSTs to http://127.0.0.1:5001/shutdown on still-alive Waitress server
3. Waitress handles it, sets _server_stopping = True, does BLE cleanup, returns response
4. Browser's next /stats poll (within ~1.5s) receives stopping: true and displays shutdown message
5. After 5s, Waitress exits via os._exit(0)
6. run.py waits 6s then calls terminate() as safety net