Support `response_format: json_object` in OpenAI server

[simon-mo]

We just merged the support for structured generation support with Outlines. The next step is to integreate with Grammar based finite state machine dottxt-ai/outlines#541 into vLLM to support arbitrary JSON format.

[interstellarninja]

is it possible to introduce a JSON mode chat template with a specific system prompt for response_format?

for eg. Hermes Pro models support JSON mode through a system prompt as follows;

You are a helpful assistant that answers in JSON. Here's the json schema you must adhere to:\n<schema>\n{pydantic_schema}\n</schema>