- Suppress ConnectionError log noise from in-flight requests draining
after a session close (poll_scheduler + keepalive log at DEBUG).
- Null KeepaliveTask.on_unreachable during a forced reconnect so the
dying session's keepalive can't flip HA availability offline after
the new session is already healthy. Ported into PR #5's
_maybe_force_reconnect flow.
Drops the _session_stop / publish-health force-close mechanism from
the original PR — PR #5's last_success_ts + _maybe_force_reconnect
already covers the 'session dead, restart it' goal via a different
path, and running both means two paths force-closing the same session
on the same failure.