feat: Add option to pass custom kwargs to litellm.completion

[ludwig-n]

Summary of PR

Adds an option to LLMConfig that allows the user to pass custom kwargs to litellm.completion.

This is useful to pass in parameters that OpenHands doesn't have a dedicated setting for, such as:

• "chat_template_kwargs": {"enable_thinking": False} to disable "thinking mode" for the Qwen3 series of models
• some sampling parameters, e.g. repetition_penalty, min_p

Example of usage in config.toml:

[llm.qwen3]
completion_kwargs.chat_template_kwargs.enable_thinking = false

Change Type

• New feature

Checklist

• I have read and reviewed the code and I understand what the code is doing.
• I have tested the code to the best of my ability and ensured it works as expected.

Release Notes

• Include this change in the Release Notes.

[raymyers]

Added a test, looks good thanks!