Do not let a console message kill the radio model

[A13xB0]

The radio model treated any engine message it did not recognise as a desynchronised stream and exited. kConsoleIn (0x06) is sent to every node the moment anything touches the fleet console, so an emulated node died seconds after boot and reported:

ERROR: radio init failed: -2

which is RadioLib for chip not found. That points squarely at wiring — pins, SPI controller, chip select — and none of it was wrong. What had happened was:

radioserver: unknown engine message 0x06
radioserver: the engine went away
radioserver: 0 transactions, 0 bytes

The chip was fine; the model had gone.

Console input belongs to the firmware's serial port, and an emulated node's serial port is the emulator's rather than this socket, so ignoring it is the whole handling. What mattered was that it not be fatal.

Unknown kinds are now skipped rather than fatal. The framing is length-prefixed and the payload has already been read before dispatch, so an unrecognised message cannot desynchronise the stream — whereas exiting takes the node down for a message it did not need. The old comment claiming otherwise was wrong about its own protocol.

This file existed only as a working copy and had never been committed; it lands here with the fix.

Verified in the application, not just in a test: an emulated Generic_E22_sx1262 running published v1.17.0 on the ScotMesh CAD study had its advert decoded by 38 nodes, from Blantyre02 at 40.7 dB down to Bishop Hill at -8.9 dB.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TGXGCDQQHLL9Bqo9GBsyrw

[A13xB0]

Obsolete: this fixes bridge/radioserver.cpp, and radioserver was retired in #18 - the emulators hold the chip model now, linked into the process rather than served from a separate binary. The console and the radio model are separate paths in the current architecture, so the failure this addressed is not reachable there. Closing rather than rebasing onto a file that no longer exists.