v0.10.1 — Gree broadcast fallback

What's fixed

The Gree AC was unreachable from the k8s deployment. This unit's firmware (V1.2.1) never answers a bind sent to its unicast address — it only responds to packets addressed to the LAN broadcast. The client sent the bind handshake and every status/command to the unit's unicast address, so the AC stayed permanently offline even though it answered the discovery scan.

Changes

• Broadcast bind fallback: try the unicast bind first (most units answer it), fall back to the broadcast address on a timeout, and remember which mode the unit wants so status/commands keep going the same way.
• Self-healing mode: if a broadcast-only unit is later replaced by one that answers unicast, the client re-learns the mode instead of staying stuck on broadcast.
• Faster reconnects: a known broadcast-only unit skips the wasted unicast attempt.
• Handshake timeout: the discovery+bind handshake now has its own budget so a cold start isn't truncated by the short status-read context, and the UDP reads honour context cancellation.
• Tests: hermetic broadcast-only unit test plus a context-semantics test; verified against the real AC (online, mode cool, set 24°C).

Rollout

Rebuild/roll the monitor deployment to pick up ghcr.io/khaledez/camonitor:0.10.1 (pinned image bump follows in a separate commit).