fix(codex): forward output token limit

[PsiACE]

Rationale

Bub passes its configured max_tokens through any-llm's Chat Completions interface. The Codex OAuth provider translates that request to the Responses API, but previously omitted the corresponding max_output_tokens field. As a result, output token limits worked with other APIs but were silently ignored for Codex OAuth requests.

Summary

• map max_completion_tokens, or max_tokens as a fallback, to Responses max_output_tokens
• preserve max_completion_tokens precedence when both values are provided
• add regression coverage for the translated payload and precedence behavior

User-facing changes

Configured output token limits now apply to Codex OAuth completions. There are no API or configuration changes.

Breaking changes

None.

Validation

• uv lock --locked
• uv run prek run -a
• uv run mypy src
• uv run python -m pytest --doctest-modules -q (269 passed)