typing: Replace mypy with ty

[PierreQuinton]

Typing changes (necessary to run ty without error):

• Use positional-only arguments for methods from classes whose subclasses rename the parameters. This includes Weighting.forward, JacobianComputer._compute_jacobian, Transform.__call__ and Differentiate.__differentiate__. This fixes a break of LSP.
• Remove now-useless typing ignore statement in MGDA
• Add necessary casts in NashMTL
• Add ignore statement for subclasses of autograd.Functions
• Fix name of parameters of methods in OrderedSet
• Make ModuleFactory generic
• Add some missing casts to PSDMatrix
• Add ignore statement when calling .grad of BatchedTensor
• Ignore type errors in the lightning example's test

Structural changes:

• Add ty check dependency
• Remove mypy check dependency
• Change section about type checking in CONTRIBUTING.md
• Remove mypy badge
• Add ty tool section in pyproject.toml
• Change CI to run ty instead of mypy
• Check typing in tests too

Need to add it in the project.toml and in the ci. Apparetnly ty is part of ruff, we may want to switch the linter too. (the astral product are fast, I'm still a bit scared of getting lockin their product though).

Fixes #550

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/torchjd/aggregation/_aligned_mtl.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_cagrad.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_constant.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_dualproj.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_imtl_g.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_krum.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_mean.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_mgda.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_nash_mtl.py	100.00% <100.00%> (ø)
src/torchjd/aggregation/_pcgrad.py	100.00% <100.00%> (ø)
... and 16 more

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ValerianRey]

FYI making some param positional-only in a public interface is a breaking change. But let's go for it. I think I really like this way of fixing the name change in override problem.
Just need a changelog entry explaining how to transition for people using e.g. weighting(gramian=gramian)

[PierreQuinton]

FYI making some param positional-only in a public interface is a breaking change. But let's go for it. I think I really like this way of fixing the name change in override problem. Just need a changelog entry explaining how to transition for people using e.g. weighting(gramian=gramian)

Yeah, realized that later. I think if we do that, we should consider all public functions and double check what we want them to be exactly in terms of positional vs named. That may go in another PR then.

[ValerianRey]

Actually not a breaking change because the change is made to forward (which shouldn't be called directly by users and isn't documented) and not to __call__.

[ValerianRey]

LGTM. This is great actually, and I like all of those changes.

We may still want to configure ty to be more strict, or to fix the type errors in the tests (we can just use the # ignore for some of them if they're boring to fix, in particular if we're testing something that's intentionally messing with types to simulate a user's mistake).

[ValerianRey]

@PierreQuinton I made quite a few changes. In particular, I fixed typing in tests and made the ci also check typing of tests (it was quite easy actually, and it could really protect us from user-facing typing problems, which may be invisible if we don't type-check tests).

[PierreQuinton]

I think LGTM, I think we can merge.