fix(openai_models): Use correct options for o3 and o3-mini models

[BillionClaw]

Summary

Fixes #731

The o3 and o3-mini models don't support temperature, max_tokens, top_p, frequency_penalty, and presence_penalty parameters. When these parameters are passed to the OpenAI API, it returns an error like:

Unsupported parameter: 'temperature' is not supported with this model.

This PR creates a new OptionsForO3 class that only includes the supported parameters:

• max_completion_tokens - Maximum number of tokens to generate (including reasoning tokens)
• reasoning_effort - Constraints effort on reasoning (low, medium, high)

Changes

• Added OptionsForO3 class with only max_completion_tokens and reasoning_effort options
• Added is_o3 parameter to _Shared class to distinguish o3 models from other reasoning models
• Updated o3 and o3-mini model registrations to use is_o3=True instead of reasoning=True

Testing

• Verified that o3 and o3-mini models now only expose max_completion_tokens and reasoning_effort options
• Verified that other reasoning models (like o1) still use OptionsForReasoning with all standard options
• Ran existing tests to ensure no regressions