fix: suppress unsupported temperature for Codex mini

[rodboev]

Codex mini models routed through LiteLLM can reject the temperature parameter. Aider already has a per-model use_temperature setting, but Codex mini currently falls through with the default request shape and sends temperature=0.

This uses the existing model-setting path to mark Codex mini as use_temperature=False, while leaving the centralized send_completion() request builder and normal temperature behavior intact.

Closes #4039

The follow-up question from coogle shaped the scope here: disable unsupported LiteLLM parameters for affected models without changing unrelated request construction.

Tests:

• rtk python -m pytest tests/basic/test_models.py -v
• rtk pre-commit run --all-files

Latest checks:

• rtk python -m pytest tests/basic/test_models.py::TestModels::test_codex_mini_send_completion_omits_temperature tests/basic/test_models.py::TestModels::test_codex_mini_sets_use_temperature_false -v
  • Passed (2/2). Both plain and provider-prefixed Codex mini variants still resolve use_temperature=False and omit temperature in send_completion.
• rtk python -m pytest tests/basic/test_models.py -v
  • Failed in this environment. litellm import crashes during aider.llm lazy load with AttributeError: module aiohttp has no attribute ConnectionTimeoutError, which affects existing constructor-heavy tests outside this change.
• rtk pre-commit run --all-files
  • Initially failed because black reformatted tests/basic/test_models.py.
  • Re-run passed after formatting.