feat: Switch from LiteLLM Proxy to Any-LLM Gateway

[njbrake]

1. Switched over to any-llm-gateway instead of litellm and updated routes and tests
2. Created script to aid auto-generation of redoc: maybe a candidate to be added as a github action workflow, but I didn't want to get too far ahead of myself 😆

Things I'm still solidifying:

1. Need to make sure that any-llm-gateway is correctly tracking VertexAI token count in streaming mode: OpenAI Chat Completions API uses an {'include_usage': true} to include the total token count from the stream at the end of the call, but I'm not sure
2. I'm not sure if you're supporting the openai model in MLPA, The litellm config seems to indicate it should look for the api key as an ENV Var, but the docker-compose isn't passing the OPENAI_API_KEY into the container so not sure if it's legacy. I don't have that configured in the gateway config and I need to make sure that the config yaml <-> env var coordination works correctly in the gateway.
3. Instead of directly running SQL queries to create the 1M users needed for testing, I run a script that exercises the any-llm-gateway user creation API. It likely makes that process take much longer, probably needs a bit more work..

Interested in your thoughts, thanks!