feat(openai): add per-request reasoning_effort override support (#617)

Add ability to dynamically control reasoning_effort parameter on a
per-request basis for OpenAI reasoning models (o1/o3/o4/gpt-5 series).

Changes:
- Add reasoning_effort field to RequestParams with Literal type hints
- Support 'none', 'low', 'medium', 'high' values in config and runtime
- Implement fallback logic: request param -> config -> default ('medium')
- Add comprehensive unit tests for reasoning_effort behavior
- Update example to demonstrate gpt-5.1 with reasoning_effort override
- Update JSON schema to include 'none' value

The reasoning_effort parameter is OpenAI-specific and only applies to
reasoning models. Non-reasoning models ignore this parameter.

Tests: Added 5 new test cases covering all reasoning_effort scenarios

Co-authored-by: dragon1086@naver.com <1gmrfyd2@>