[codex] Update reasoning effort metadata

[Kamilbenkirane]

Summary

Updates text model reasoning controls to match provider effort settings:

• Switches claude-sonnet-4-6 from numeric thinking budget to named thinking levels.
• Adds xAI reasoning effort choices for grok-4.3.
• Adds Mistral reasoning effort choices for mistral-medium-3-5 and mistral-small-2603.
• Maps Mistral string reasoning choices to reasoning_effort while keeping numeric Magistral budget behavior on prompt_mode.

Validation

Pre-push hooks passed:

• Ruff lint and format
• mypy for source and tests
• Bandit
• unit tests with coverage

Note: the first push attempt exposed a missing optional google-auth dependency in the fresh worktree venv; syncing the repo's gcp extra fixed the local test environment, with no tracked file changes.