Add the ability to specify a fallback language model

[neubig]

Once we have LLM profiles implemented: #1882

we should add the ability to specify a primary LLM profile and optional fallback profiles to query when LLM calls to the primary profile fail.

This will allow us to continue having the agent work when a particular LLM provider goes down.

[raymyers]

Linking LiteLLM docs on fallbacks for reference on options.

https://docs.litellm.ai/docs/proxy/reliability