feat(config): unify reasoning effort defaults

[xiami762]

Summary

• Unify configurable reasoning effort defaults across supported models.
• Keep the change focused on reasoning effort behavior only.

Key Changes

• Add a global default_models.default_parameters.reasoning_effort setting with a default of high.
• Resolve model-specific reasoning effort overrides on top of the global default.
• Use the configured effort for OpenAI reasoning models and align Kimi K3 with the shared default.
• Exclude default_parameters from the existing default-model listing API.

Impact Scope

• User-visible behavior: OpenAI reasoning models and Kimi K3 now default to high effort.
• Compatibility / migration: Existing model-level reasoning_effort settings continue to override the global default; no migration is required.
• Configuration / environment: Adds an optional global reasoning effort default.
• Dependencies: None.
• Performance / resources: No material impact.
• Security / permissions: No impact.

Business Logic to Review

• Precedence between global default parameters and model-specific settings.
• Provider option mapping for OpenAI and Kimi K3.

Why This Approach

• Reuses the existing model settings structure and adds only the minimum global-default lookup needed to keep provider behavior consistent.

Test Plan

• uv run pytest tests/config/test_config_writer.py tests/provider/test_provider_options.py tests/provider/test_thinking_params.py tests/provider/test_model_management_p2p3.py
• uv run ruff check flocks/config/config_writer.py flocks/provider/model_manager.py flocks/provider/options.py tests/config/test_config_writer.py tests/provider/test_provider_options.py tests/provider/test_thinking_params.py
• git diff --check

Compatibility, Migration & Rollback

• No breaking configuration migration. Existing per-model values take precedence.
• Rollback by reverting this PR.