fix: always include max_completion_tokens for OpenAI compatible providers

[roomote]

This PR fixes an issue where the response context limit was randomly ignored when using OpenAI Compatible API providers like koboldcpp.

Problem

When using OpenAI Compatible API providers, the max_completion_tokens parameter was only included when includeMaxTokens was explicitly set to true. When undefined, the parameter was omitted, causing providers like koboldcpp to fall back to their default limit (512 tokens), which is insufficient for generating complete diffs.

Solution

• Changed the default behavior to include max_completion_tokens when includeMaxTokens is undefined (treating undefined as true for backward compatibility)
• Added validation to only include positive max token values (avoiding invalid -1 values)
• Updated tests to reflect the new default behavior
• Added specific tests for OpenAI compatible provider scenarios

Testing

• All existing tests pass
• Added new tests to verify the fix works correctly
• Tested with both includeMaxTokens: false (explicitly disabled) and undefined (default enabled)

Fixes #6265

---

Important

Ensure max_completion_tokens is always included for OpenAI compatible providers unless explicitly disabled, with updated tests in openai.spec.ts.

• Behavior:
  • Always include max_completion_tokens for OpenAI compatible providers unless includeMaxTokens is explicitly false in openai.ts.
  • Treat includeMaxTokens as true when undefined for backward compatibility.
  • Validate to include only positive max_completion_tokens values.
• Testing:
  • Updated tests in openai.spec.ts to reflect new default behavior.
  • Added tests for scenarios with OpenAI compatible providers, including when includeMaxTokens is false or undefined.
• Misc:
  • Minor updates to comments and logic in addMaxTokensIfNeeded() in openai.ts.

^{This description was created by for e83f5cc. You can customize this summary. It will automatically update as commits are pushed.}

[daniel-lxs]

Closing for now, I don't think this is the right solution