While trying to use the types.ModelConfiguration data to configure completion providers, it became clear I missed something.

We expose the types.GenericProviderConfig as a bare-bones way to express LLM provider configuration, and it does great. But just knowing the (endpoint, access token) isn't enough to actually serve requests.

We need to know the shape of API requests. e.g. if we should pass that endpoint/access to an OpenAI-compatible, Fireworks-compatible, or Google-compatible API provider.

NOTE: An alternative approach would be to not add this field to GenericProviderConfig entirely, and instead add AnthropicProvider *GenericProviderConfig, GoogleProvider *GenericProviderConfig, etc. to the ServerSideProviderConfig. And essentially have the "service name" be implicit based on which field of ServerSideProviderConfig you found the object.

There are advantages to that approach, but this seemed a bit cleaner IMHO. WDYT?

Test plan

Added new unit tests

Changelog

NA