fix: decode base64 embeddings as little-endian floats

[HostX0]

• I understand that this repository is auto-generated and my pull request may not be merged

Root cause

The endpoint's base64 embedding representation contains little-endian float32 bytes, but the handwritten response parser used native byte order in both its NumPy and stdlib paths. On big-endian Python platforms, that silently produces incorrect vector values. The existing fixture was also native-endian, so it followed the host instead of the wire format and could not expose the defect.

Fix

Decode the NumPy path with the explicit <f4 dtype, and byte-swap the stdlib array("f") path on big-endian hosts. The fixture now constructs little-endian wire bytes explicitly, with a regression test covering the stdlib byte-swap path.

Tests

• python -m pytest -q -n 0 tests/lib/test_embeddings.py — 58 passed with Pydantic v2
• the same test file under the Pydantic v1 compatibility environment — 58 passed
• targeted Ruff format/check and Pyright checks passed
• targeted Mypy check for src/openai/lib/_parsing/_embeddings.py passed

The repository-wide lint script was also attempted. Ruff passed, while Pyright stopped on 17 pre-existing errors in unmodified audio helper and UV workflow test files.