Add unit tests and a GitHub Actions workflow to run them

[ivnpsvc]

Following up on #28 and the discussion with Zhongqi on Discord, this adds a first set of tests and a workflow that runs them on pull requests and on pushes to main.

What's included

• tests/: 33 unit tests that run in a few seconds and don't download any model weights (they also pass offline). They use small generated images and hand-built detection results:
  • the package and its main modules can be imported
  • MegaDetectorV6: an unknown version raises a clear error, the default version is valid, class names
  • utils.post_process: detection JSON (contents, excluded categories, relative paths), Timelapse JSON ([x, y, width, height] boxes, categories, max_detection_conf), sorting into Animal/ and No_animal/ (including the threshold), crops (names and sizes), annotated images, and keeping camera subfolders with input_dir
  • data: image file extensions, letterbox (size, aspect ratio, padding) and the transforms
• .github/workflows/tests.yml: runs the tests on Python 3.10 and 3.12, with the actions pinned to the same commit SHAs as publish.yml.

Note on dependencies

The workflow installs soundfile, librosa and setuptools<82 explicitly, because import PytorchWildlife fails without them (#26). I left pyproject.toml unchanged so that fix can be handled in #26; once it's merged, these can be removed from the workflow.

Running locally

pip install -e . pytest soundfile librosa "setuptools<82"
pytest

[ivnpsvc]

@microsoft-github-policy-service agree