add github workflow using pip installation for python 3.8 - 3.12

[gnzng]

This PR adds a github workflow for setting up the environment as well as running the pytest for every push and pull into master from python 3.8 - 3.12. Unfortunatelly, 3.7 is not widely supported (anymore).

It is all running on pip, which I think is a better way forward, than using conda. pytorch doesn't even support installation via conda anymore: https://pytorch.org/get-started/locally/. If that is good way forward we could probably change the docs and update the installation.

[allevitan]

This looks good, but I want to check that the updated requirements.txt file works with the existing installation instructions.

Actually, this might be a good time to update the documentation for installation, now that pytorch no longer supports conda. I'd like to pull these two changes in together, so that the documentation stays current with the codebase.

[allevitan]

Haha, apparently I should not try to review a code change so soon after waking up. Let me revise what I said to "seems perfect, but could you update the installation info in the docs and add it to this pull request?"

If you don't have the docs set up, feel free to edit in markdown and add it to the PR - I can test building the docs and then push a final version.

[gnzng]

Just opened another PR over at cdtools-developers/cdtools-docs#1

[gnzng]

I added a demonstration on how that would look like in a PR over in my own cdtools fork: gnzng#1

[allevitan]

Perhaps we don't need to install with -e (which, I believe will go away eventually), and we might not even need to to do the pre-install of requirements.txt. But this script should work, no need to change.

[gnzng]

I do agree. I just wanted to implement an example, which is very close to your current installation workflow. In general I see a lot installations using the pyproject.toml, which allow you to install the package with a certain subset of packages, eg. pip install .[docs]. Here is an example on an unreleased project build on this cookiecutter skeleton: https://github.com/audreyfeldroy/cookiecutter-pypackage

dependencies = [
  "matplotlib",
  "numpy",
  "torch",
  "loguru",
  "h5py"
]

[project.optional-dependencies]
docs = [
    "sphinx",
    "sphinx-markdown-tables", 
    "numpydoc",
    "sphinx_copybutton",
    "myst_parser",
    "sphinx_rtd_theme",
    "sphinx_rtd_dark_mode"
]
tests = [
    "coverage",
    "coveralls",
    "codecov",
    "pylint",
    "pytest-cov"
]
dev = [
    "mypy",
    "pytest",
    "ruff"
]

[allevitan]

I think this all looks good and is a clear improvement. Let's set it up and see what happens.