Small items from the review of 1.0.5

[DAB-LABS]

The sixth review found nothing wrong in the code, so this is a tidy-up round. The one behaviour change is small: if the device says the session key has expired and the library's attempt to log in again then times out, the caller now sees the timeout instead of the device's original message, which is what actually happened.

The rest is a Windows safeguard for the discovery socket, an honest rewording of the RM Max note (the mapping comes from an upstream pull request whose testers reported the device answering "locked", so it is listed as reported rather than confirmed), a log level, and one loop closed the same way as its sibling.

Tested on Python 3.13 and 3.14, 274 tests, plus a live run against an RM4 Pro.

Coding assistant's notes on this change:

NetworkTimeoutError and EndpointClosedError raised during the library's own re-authentication now propagate; other BroadlinkExceptions still return the request's original reply as before, so a locked device still surfaces as AuthorizationError for the caller's own reauth flow. _open_endpoint binds to ("0.0.0.0", 0) when neither address is given, for the proactor loop on Windows. The RM Max comment and the 1.0.0 changelog entry say only what upstream mjg59#838 showed. The undecodable-capture message is debug rather than warning, since the window re-arms by itself, and _capture_rf_loop closes its inner generator with aclosing like the IR loop. Typing, py.typed and the CLI move remain deferred on purpose.