chore: migrate to hatch + use hatchling for the build backend, drop Python 2 support

[agriyakhetarpal]

Description

• Stemmed off from Fix numpy v2 breaking changes #618 – the packaging infrastructure for autograd has been migrated to use hatchling.build instead of setuptools.build_meta as the build backend. Some missing metadata entries have been added in pyproject.toml.
• Dependence on the six and future libraries has been removed, since all of the code is compatible with Python 3.
• Python 2 support has been dropped and the wheels are tagged with py3 for the Python tag instead of py2-py3.
• Lastly, I added some future TODOs so that I can keep track of them sometime soon.

What issue does this PR close/address?

Closes #606
Closes #513, since we won't need setuptools-specific files anymore.

Caveats (i.e., tasks for future PRs)

• to enable VCS versioning via dynamic = ["version"] and the hatch-vcs plugin to derive the version from Git; and
• to use hatch environment commands in place of tox through the addition of a [tool.hatch.run] table in pyproject.toml.

Additional notes

The conda.yaml file doesn't look like it's going to work anyway; I'll modify this in another PR and switch to an environment.yaml file if needed, or we can drop it sometime soon: all of our dependencies both for regular usage and development work are available via PyPI.

[agriyakhetarpal]

I'm hitting the same error as the other PR, which suggests that the problem is with either Hatch or PyPy-3.10-Windows – changing the build backend did not work. Either way, I'm sure these changes are beneficial. I guess it should be safe to drop this platform and test just PyPy-3.10-Ubuntu and PyPy-3.10-macOS. @fjosw, what do you think?

[agriyakhetarpal]

Similar error: actions/runner-images#5009 – I bumped setup-python to v5 and then to v5.1.1; that doesn't resolve it either. It could be a bug in PyPy itself since 322122547 means it's trying to access an invalid memory address.

[agriyakhetarpal]

Looking at the logs, the installation time for all platforms on PyPy is unusually high. It shouldn't take thirteen minutes just to install a pure Python project at all.

[fjosw]

Hey @agriyakhetarpal, I would be fine with not running PyPy tests on windows, or even dropping all PyPy tests if they cause so many troubles, whatever you prefer.

[agriyakhetarpal]

Sure, I dropped it for just Windows for now, since it's tested elsewhere and PyPy 3.9 is passing for Windows. I have a branch locally for switching to nox because it will be easier for us to modify the testing infrastructure and bring it closer to the Scientific Python ecosystem's guidelines – but I don't wish to hold up your PR at this time, so this is ready for review.

[agriyakhetarpal]

Lots of potential improvements and maintenance to do, I'm excited 😄

[agriyakhetarpal]

Just realised I pushed the previous commit to master when I meant to do so here – we should set up branch protections.

[agriyakhetarpal]

I don't have access to change branch protection rules in the settings. @j-towns, could you please help us out with this? Also, since not many PRs are active, we could additionally look at renaming master to main.

[j-towns]

Yep, since they're not urgent I'll look at those things in around 10 days when I'm back from holiday.

[agriyakhetarpal]

Sure, thanks! Enjoy your holiday!

[agriyakhetarpal]

@fjosw, please let me know if this is good to go for you – I'm awaiting your review :) I think it is, at least from my end.

[fjosw]

Hi @agriyakhetarpal, sorry for the delay, your changes look good to me! I guess "numpy<2" is only temporary to let the tests pass?

[agriyakhetarpal]

Thanks for the review, and no worries about the delay! Yes, the upper pin is just temporary – just updated your branch in the other PR where the pin doesn't exist 😄