Fix import, argument logic and model mismatch

[scris]

Thanks for your project, which is very helpful.

However, I found some bugs in the published code, and they're resolved by this PR.

This PR solves:

1. Some import issues found in evaluate/*.
2. Typo and sub-menu issues found in the logic of argument list in main.py.
3. Backend mismatch between OpenAI and Gemini in evaluate/providers/__init__.py.

Maybe you can look through this PR. Thanks a lot.

---

By the way: Can you publish the id list of all-agree problem sets? Thank you very much.

[cr-gpt]

Seems you are using me but didn't get OPENAI_API_KEY seted in Variables/Secrets for this repo. you could follow readme for more information