tests: test_env_override_wins hardcodes Linux ffmpeg path on macOS

[promiseeuler]

Summary

While validating PR #107 on macOS, I found that test_env_override_wins hardcodes /usr/bin/ffmpeg. The implementation correctly rejects nonexistent override paths, so the test fails on supported macOS installations where FFmpeg is normally elsewhere (for example, Homebrew uses /opt/homebrew/bin/ffmpeg on Apple Silicon).

Environment

• HFlow upstream main (verified against the current raw source on 2026-08-22)
• macOS, Apple Silicon
• Python 3.14.2
• uv 0.12.5
• FFmpeg installed and available on PATH, but not at /usr/bin/ffmpeg

Steps to reproduce

uv sync --locked --all-extras
uv run pytest -q tests/test_ffmpeg.py::test_env_override_wins

Expected behavior

The platform-supported test should verify that an existing override path wins without depending on a Linux-specific filesystem location.

Actual behavior

E hflow.ffmpeg._binary.FfmpegNotFoundError:
E HFLOW_FFMPEG=/usr/bin/ffmpeg does not exist
1 failed

The failure is deterministic because ffmpeg_path() intentionally checks Path(override).is_file().

Possible cause

tests/test_ffmpeg.py:67 sets:

monkeypatch.setenv(FFMPEG_ENV_VAR, "/usr/bin/ffmpeg")

That path is not portable to macOS, despite CONTRIBUTING.md listing macOS as a supported native development platform.

Proposed solution

Create a temporary file with tmp_path, set HFLOW_FFMPEG to that path, and assert the resolver returns it. The test only exercises path precedence, so it does not need a real executable:

def test_env_override_wins(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
    cleared_binary_caches: None,
) -> None:
    override = tmp_path / "ffmpeg"
    override.touch()
    monkeypatch.setenv(FFMPEG_ENV_VAR, str(override))
    assert ffmpeg_path() == override

This keeps the test hermetic and platform-independent.

[WilliamK112]

I'd be happy to take this. I'll keep the change scoped to making test_env_override_wins hermetic with tmp_path, then run the focused test and the repository quality gates before opening a PR. I confirmed the issue is unassigned and there is no overlapping open PR. AI assistance disclosure: I am using OpenAI Codex to help inspect, implement, and validate the change, and I will review the final diff and evidence.