Missing OpenAI functionality

[Tostino]

Just wanted to mention that I am looking at the vectorizor branch, as that seems to be the active development branch as far as I can tell.

The current OpenAI API is not really compliant with the python version of the API, restricting it's use.

The extension is missing a bunch of arguments for the openai endpoints, and not all of the arguments that are there behave in-line with the API spec. Some of the missing arguments are needed for using proxy solutions or custom samplers on open source inference servers for example, The return types for the embed endpoints also seems like it should just be json, because there are options to return the data in either base64 or floating point numbers for example.

I think the functions that provide a "simple" wrapper should be designed on top of the raw functions that return json rather than strict data types. Probably in a bit more structured a way than the current "simple" chat completion function. I haven't gotten to this yet. Having a full featured base set of functions was my priority first.

Either way, I put in a good bit of work into a branch, but cannot get the test suite to pass due to white-space issues with the expected files. I'm at my wits end with messing with the test suite considering I cannot run it locally because your build scripts aren't compatible with podman and that's what I use on my dev machine.

Comparison with branch: https://github.com/timescale/pgai/compare/vectorizer...Tostino:pgai:fix-openai-api?expand=1

The last major argument I haven't taken into consideration here is timeout which will need a little more thought because it needs to be supported at both the client level as well as the function level. Was planning on doing that after getting some feedback.