Support multiple competing LLMs

[gbenson]

e.g.

• main/primary (x86+GPU, larger model)
• backup/secondary/fallback (arm64, some smollm2, in case network down or ollama not running)

How to prioritize?

• chat-router knows & invokes each directly (easy with current rpc-invocation style, easy to experiment (don't need to change hive-ollama), prob only good for main+fallback)
• hive-ollamas consume fanout & indicate their model's "goodness" (probably a better solution definitely harder to experiment since both chat-router and hive-ollama need synchronous changes, could add external (paid) options maybe a later enhancement)

[gbenson]

This might be easier with the hive-messaging reorg...