fix: coerce exception code to string or None

[Rajeev91691]

Coerce error code inside APIError body to str | None to prevent validation errors with construct_type when the API returns integer codes (e.g. 400). Added unit test in tests/test_client.py.

[ting-hong-shieh]

Reviewed b00a48a against #3531. The focused regression passes, and the normalization keeps the public str | None contract intact. One reproducible formatting failure remains below. No API call or external service was used.

Validation snapshot:

• Focused pytest: 1 passed, 176 deselected.
• Ruff 0.14.7 check: passed.
• Ruff 0.14.7 format check: failed because tests/test_client.py would be reformatted.
• git diff --check: failed at tests/test_client.py:2971.

[ting-hong-shieh]

Please remove this extra blank line at EOF. Using the repository-pinned Ruff 0.14.7, ruff format --check src/openai/_exceptions.py tests/test_client.py reports that this file would be reformatted; git diff --check independently reports tests/test_client.py:2971: new blank line at EOF. The focused regression itself passes (1 passed, 176 deselected).

[CacinieP]

Confirmed the remaining formatting failure — it's a single extra blank line at end of file, introduced by the new test:

$ ruff format --check src tests   # ruff 0.14.7, as in CI
Would reformat: tests/test_client.py

$ git diff --check
tests/test_client.py:2971: new blank line at EOF.

tests/test_client.py on this branch ends with assert err.code is None\n\n (two trailing newlines) while every other file ends with exactly one. Deleting the final blank line clears both the ruff format check and the git diff --check failure — the rest of the diff (src and tests) is otherwise clean under ruff format --check and ruff check.

One-liner fix:

printf '%s\n' "$(cat tests/test_client.py)" > tests/test_client.py

or just remove the last empty line in your editor and push.