fix: resolve moonshotai/kimi-k2-instruct max_tokens limit issue

[roomote]

This PR fixes the issue where the moonshotai/kimi-k2-instruct model was causing 400 errors due to max_tokens exceeding the API limit.

Changes Made

1. Updated model configuration: Changed moonshotai/kimi-k2-instruct maxTokens from 8192 to 16384 to match the actual API limit
2. Added validation in BaseOpenAiCompatibleProvider: Ensures that max_tokens never exceeds the model's configured limit, even when users override with modelMaxTokens
3. Fixed TypeScript types: Added proper type checking to handle nullish values for maxTokens

Problem Solved

• Prevents 400 errors: "max_tokens must be less than or equal to 16384"
• Ensures user-configured modelMaxTokens values are capped at the model's actual API limits
• Maintains backward compatibility while preventing API errors

Testing

• All existing tests pass
• TypeScript compilation successful
• Linting passes

Fixes #5729

---

Important

Fixes max_tokens limit issue for moonshotai/kimi-k2-instruct model by updating configuration and adding validation in BaseOpenAiCompatibleProvider.

• Behavior:
  • Updated maxTokens for moonshotai/kimi-k2-instruct in groq.ts from 8192 to 16384.
  • Added validation in BaseOpenAiCompatibleProvider to cap max_tokens at model's API limit.
• TypeScript:
  • Improved type checking for maxTokens to handle nullish values.
• Testing:
  • All existing tests pass.
  • TypeScript compilation and linting successful.

^{This description was created by for ebd6aec. You can customize this summary. It will automatically update as commits are pushed.}

[valuin]

+1 facing this issue ty for the PR

[mrubens]

Fixed in another way