Add timeout param to all chat_async methods, add backup model param

[rishsriv]

Context

We sometimes have scenarios where the LLM we are using goes down sporadically, or just has an insanely long response time. For example, sonnet will (in about 1 in 200 requests) take 60+ to generate a response that it usually does in 5 seconds.

Solution

We want to:

1. Add a timeout parameter to our LLM models (the OpenAI and Anthropic SDKs support this, Google's python-genai does not yet from what I can tell)
2. If the original model either fails or does not yield a response in the desired period, then use the backup model

This PR implements this solution.