align reasoning model params with OpenAI spec & add 'minimal' effort support

[RaajeevChandran]

Summary

Fixes #649 by resolving HTTP 400 error when using OpenAI's reasoning models (o1 and gpt-5-nano) by aligning the request payload with the OpenAI API specification referred from the official OpenAI community

Changes

• Implemented the new reasoning_effort parameter for openai.com host (I've ensured backward compatibility fr other providers)
• Intentionally omit temperature and top_p from the request when an OpenAI reasoning model is detected. Looks like these parameters are forbidden by OpenAI when reasoning features are active.

• I've also created a new "minimal" effort level in the OpenAIReasoningProfile for reasoning models.

• Behavior change

• UI change (screenshots below)

• Refactor / chore

• Tests

• Docs

Test Plan

• Connect to the OpenAI provider and verify that models like o1-preview or gpt-5-nano no longer return "Unknown parameter" errors.
• Verify that non-OpenAI providers still receive the legacy reasoning object
• Open the model settings for a reasoning model and verify that "minimal" is now a selectable option in the segmented control

Checklist

• I have read CONTRIBUTING.md
• I added/updated tests where reasonable
• I updated docs/README as needed
• I verified build on macOS with Xcode 16.4+