[Feature] Adds basic support for image content in OpenAI chat routes

[fozziethebeat]

This adds piping so that OpenAI Chat Requests can forward input images to be properly processed by LLaVa type models. This includes a bit of piping, expanded OpenAI request types (copied from openai-python) and some testing.

I've tested this manually by running:

python3 -m sglang.launch_server --model-path liuhaotian/llava-v1.5-7b --tokenizer-path llava-hf/llava-1.5-7b-hf --chat-template vicuna_v1.1 --port 30000

and

python test/srt/test_openai_server.py

Output came out as expected:

 Paris. The country is known for its rich history, art, culture, and cuisine. It is also home to the
 United Nations Educational,
 Paris. The country is known for its rich history, art, culture, and cuisine. It is also home to the
 United Nations Educational,
 The capital of France is Paris. (pronounced "Par-ee")
 The image displays a graph with various data points, including a line that goes from 0 to 100.

There are several other lines on
 1. United States - Washington D.C.
2. Canada - Ottawa
3. Mexico - Mexico City

[comaniac]

LGTM. I'll leave to others for the change in tokenizer manager.

[merrymercy]

@fozziethebeat Thanks!

[fozziethebeat]

Thanks for taking the contribution!