[Frontend] Add OpenAI Vision API Support

[ywang96]

This PR is the final step of the release milestone of adding support for GPT4V compatible inference.

This PR fixes

• Support VLM model and GPT4V API #2058
• [Doc/Feature]: Llava 1.5 in OpenAI compatible server #3873
• [Usage]: Make request to LLAVA server. #4205
• [Usage]: Passing image to the vllm api endpoint #4826

and closes #4200

[ywang96]

@DarkLight1337 This PR is probably good for first pass - I will mark it as "Ready for review" after I address your first round of review, along with adding more test and documentation.

[DarkLight1337]

Thanks for the hard work! I have left a few comments.

[ywang96]

With docs and tests modified - going to make this PR ready for review cc @DarkLight1337.

[DarkLight1337]

Assuming that the tests pass, LGTM. I will double-check the docs after they have been built.

[ywang96]

Thanks for the review @simon-mo! I have addressed the comments and all related tests have passed. Could you take a look at the aiohttp implementation and let me know if there's any issue? Thanks!

[simon-mo]

• can you add the package to requirements
• can you confirm the test is ran in ci (the entrypoints tests config in pipeline is complex)

[DarkLight1337]

• can you add the package to requirements

aiohttp was already added to common requirements in #4794.

• can you confirm the test is ran in ci (the entrypoints tests config in pipeline is complex)

I have downloaded the CI log and confirm that this is the case (with the help of Ctrl+F).