fix: omit ChatGPT max output token parameter

[AlexeyMatskevich]

Summary

• Remove the unsupported max_output_tokens field from ChatGPT OAuth /responses requests.
• Keep OpenAI/Copilot provider behavior unchanged; this only affects the ChatGPT backend endpoint that rejects the parameter.

Verification

• bun run check
• Confirmed locally in OpenCode that web-search no longer fails with Unsupported parameter: max_output_tokens.

Closes #17

[emilsvennesson]

Thanks for the PR!