Fixed behaviour for refresh

Re-check connected clients immediately after a reload

A reload rebuilt the known set but did not re-evaluate anything, so a
device it had just forgotten stayed silent until the next WebSocket
event or fallback poll. With FALLBACK_INTERVAL at 600 and the device
already connected (so producing no events), that meant up to ten minutes
of apparent silence after /reload fresh discarded an allow.

Adds a dedicated recheckCh rather than reusing trigger, which runs the
WebSocket delay and IP-wait and would label the check as WebSocket
sourced. The send is non-blocking, so a reload cannot stall the loop when
a check is already queued.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QSwZJ6jnJ8Gs8mLLkx2mHY