Add assistants, threads, runs, and messages APIs (assistants beta v2)

[agrinman]

This adds support for OpenAI's Assistants API (threads, runs, messages, etc).

Note: one deviation/pattern here is I introduce a OpenAiClient which I think is a bit more ergonomic by giving users of the API a single client instead of building a new client, parsing credentials each time. I haven't rewritten the other methods for compatibility (we could add them and then deprecate the others) but I first wanted to show the approach here before going down that road!

[rellfy]

Thanks for the contribution, I'll try to have a look at the draft this weekend.

I agree that currently the approach with Credentials can seem a bit unusual. It was mostly a quick fix for allowing the library to support different inference providers at runtime. But I wouldn't say you're creating a client each time, with the current approach you simply build a request and then input the credentials which define what provider and model to use.

I haven't done any major refactors since I forked the original repo, but I'm not opposed to the idea of having a more "standard" client struct and constructing requests from there rather than passing Credentials to a request builder.