fix: clean Ctrl-C shutdown — daemon teardown + quiet SSE cancellation

[ArthurBernard]

Summary

Two related shutdown defects, both against uvicorn 0.49's signal/streaming behavior:

Defect A — start --serve loses its teardown. uvicorn 0.49's Server.serve() unconditionally wraps itself in capture_signals(), which restores the pre-serve() signal disposition and re-raises the captured SIGINT/SIGTERM with that (default) disposition right after serve() returns — killing the process before _run_daemon's own finally (scheduler shutdown + supervisor.shutdown()) can run. Fixed by overriding the built Server instance's capture_signals with a no-op context manager (there is no public install_signal_handlers config flag on this uvicorn version — confirmed it raises TypeError at construction) and installing the daemon's own SIGINT/SIGTERM handlers on the running loop around await server.serve(): first signal sets should_exit, second sets force_exit. The headless path (no --serve) is untouched.

Defect B — traceback spam on every shutdown with a connected dashboard. With a browser/curl holding /api/events, the 3s graceful-shutdown timeout force-cancels the connection and uvicorn logs a multi-screen CancelledError/anyio.WouldBlock traceback at ERROR level. Both SSE generators now catch asyncio.CancelledError around their streaming loop and end cleanly (the merged generator also cancels any outstanding per-iteration getter task). This alone isn't sufficient, though: the traceback actually originates in Starlette's own StreamingResponse.__call__ (an anyio task group racing "stream the body" vs "listen for disconnect" — the code path uvicorn 0.49 is stuck on since it declares ASGI spec_version: "2.3", below the 2.4 Starlette checks for its newer implementation), confirmed by reproducing the identical traceback against a bare FastAPI app with no custom middleware at all. So a small logging.Filter on uvicorn.error also drops the "Exception in ASGI application" record specifically when the underlying exception is a plain CancelledError — any real exception is still logged normally.

Verified (pty driver, genuine keyboard Ctrl-C, SSE client attached)

1. trading-bot dashboard + SSE + one ^C → exit 0 in ~4.2s, dashboard stopped printed, only the expected "Cancel 1 running task(s)..." line, no CancelledError traceback.
2. trading-bot start --serve --interval 3600 + SSE + one ^C → exit 0 in ~4.2s, daemon stopped (all strategies shut down) printed, no traceback.
3. trading-bot start --interval 3600 (headless) + SIGINT → unchanged, daemon stopped in ~1s.

Test plan

• python -m pytest — 1170 passed (added: single/merged SSE cancellation tests, _run_daemon signal-ownership test, logging-filter unit tests)
• ruff check trading_bot/ clean
• ruff format --check trading_bot/ clean
• mypy trading_bot/ clean
• pty-driver verification of all three scenarios above

🤖 Generated with Claude Code