Define dependencies for Python package

[berombau]

If we want to submit our package to PyPi, we need to define the requirements in a requirements.txt or in pyproject.toml.

Right now we only have 1 conda environment.yaml with everything: build-dependencies, cudatools, direct dependencies, jupyterlab, dev dependencies, test dependencies...

Conda-based plugins in napari are however supported and is also the future direction of napari packaging. But we want to support both PyPi and conda for now.

• define minimal dependencies to run napari plugin
• ensure pip install -e . still works
• also install Hydra and CLI
• document GPU installation with PyTorch
• support [testing] (CI tooling, pytest...) and [dev] (testing + jupyterlab...)
• edit GitHub Actions to build and publish to PyPI with Twine

[lopollar]

@ArneDefauw , is this fixed now?
At least for windows, right?

[ArneDefauw]

I think we still need to define minimal dependencies, if we want to publish to Pypi. I.e. dev dependencies, jupyterlab should not be in this minimal dependencies file.
I don't know if we want to implement all the features described above by @berombau .

Minimal would be to have

• define minimal dependencies to run napari plugin
• ensure pip install -e . still works
• document GPU installation with PyTorch