[codex] fix local endpoint recovery after IP changes

[copystring]

Summary

Fixes local TCP recovery when a Roborock device receives a new LAN IP address. The local IP is now handled as dynamic local transport endpoint state inside localApi, instead of being treated as a one-time startup discovery result.

Root Cause

localApi.localDevices[duid].ip was populated from startup discovery/probing and TCP reconnects kept retrying that stored IP. If DHCP moved a robot from one address to another, reconnects could keep targeting the stale address indefinitely and log repeated EHOSTUNREACH warnings.

Changes

• Keep UDP discovery as a persistent endpoint feed and merge discovered endpoints per device.
• Reset only the affected TCP socket, nonces, and pending TCP requests when a discovered endpoint changes.
• Mark unreachable TCP endpoints stale for connect-time network failures and trigger a throttled endpoint refresh instead of looping against the same stale IP.
• Add MQTT fallback endpoint refresh via get_network_info / service.get_net_info.
• Add a slow-poll background refresh for local endpoints that are known but currently disconnected, without blocking fast polling.
• Add README WORK IN PROGRESS changelog entry using the existing ioBroker schema.

Validation

• cmd /c npx vitest run src/lib/localApi.test.ts test/unit/device_online_homedata_sync.test.ts
• cmd /c npm run typecheck
• cmd /c npm run test:unit