Offload compute by allowing users to sign in with their LLM tokens

[rockett-m]

This will allow for distributed compute and no rate limiting (versus using a single API key).
If we can allow for "sign in with OpenAI/LLaMa/Cerebras/HuggingFace/Grok/others" then users can draw upon their own purchased compute plans and no bottleneck on our end calling our backend functions.