Enable OpenAI compatible endpoints

[rothgar]

I'm curious if there would be interest in accepting a PR that enables OpenAI endpoint (e.g., vllm) in addition to Gemini. I don't want to send the review without knowing if that's something this project would support or want.

//        AX_PLANNER_TYPE        - "gemini" or "openai"
//        AX_OPENAI_BASE_URL     - OpenAI-compatible endpoint
//        AX_OPENAI_MODEL        - model name
//        AX_OPENAI_TIMEOUT      - request timeout (e.g. "60s")
//        AX_OPENAI_API_KEY_ENV  - name of the env var holding the API key

I have it implemented in a local fork, but have not tested all of the demos to verify all of the functionality is there yet.

[rakyll]

We are rebasing to controller to allow bringing any harness to AX. In that case, rather than a model, we could be able to bring an OpenAI harness and should be able to support Codex. Let's catch up once we have the baseline interfaces.