Add support `extra_query` 、`extra_body` and `extra_headers ` for the custom endpoint AI model

[axetroy]

Some models require additional parameters for configuration. For example, the on/off state of the thinking mode and the intensity of the thinking mode.

Therefore, we would like to add a setting to specify additional parameters when requesting the model.

This parameter comes from the Open AI SDK: https://developers.openai.com/api/reference/python

[gpotter2]

Hi. I made a PR to take a shot at this, #329215

I've added more context in the description, but tldr; a parameter like this one would allow to support various engine-specific functionalities in "custom endpoint" scenarios. Both vLLM, SGLang, DeepSeek (the service) and QwenCloud can take advantage of this, at the very least.