Make the LLM max-token request field configurable

[hyspace]

Summary

This change makes the max-token request field configurable for the OpenAI-compatible LLM integration.

Motivation

The known failure that prompted this change was observed with Azure AI Foundry, where requests fail when the max-token field does not match the target model's API expectations.

Azure's guidance for this issue indicates that some deployments require max_completion_tokens instead of max_tokens:
https://learn.microsoft.com/en-us/answers/questions/2139738/openai-badrequesterror-error-code-400-((error-((me

Making this field configurable keeps Koe compatible with endpoints that require a different parameter name, while preserving the current default behavior.

Changes

• add a max_token_parameter config option under llm
• support max_tokens and max_completion_tokens when building chat completion requests
• keep the default value as max_tokens for backward compatibility with existing setups
• document the new option in the sample config and README

Testing

• cargo check

[hyspace]

noticed new requirements for contributing, updating PR

[missuo]

noticed new requirements for contributing, updating PR

You just need to modify the commit message, there are no other major problems.