Add unit tests for _openai_retry helpers (77% -> 95% coverage)

[jluocsa]

Summary

src/agents/models/_openai_retry.py was at 77% coverage: several low-level helpers (header parsing, retry-after parsing, status/error-code extraction) and a few get_openai_retry_advice branches were untested. This adds focused unit tests that bring the module to 95% without touching any runtime code.

Changes

• New tests/models/test_openai_retry_helpers.py covering:
  • _header_lookup for plain mappings (case-insensitive) and httpx.Headers.
  • _get_header_value via the response_headers attribute path.
  • _parse_retry_after_ms and _parse_retry_after for numeric, HTTP-date, negative, and invalid inputs.
  • _get_status_code via status_code and status attributes.
  • _get_error_code via nested and top-level body mappings.
  • get_openai_retry_advice for the unsafe_to_replay, websocket-request, x-should-retry: false, and retry-after-only branches.

Test plan

• uv run pytest tests/models/test_openai_retry_helpers.py passes (11 tests).
• Combined with the existing test_model_retry.py, _openai_retry.py coverage rises from 77% to 95%.
• ruff format, ruff check, and mypy are clean on the new file.