Isolate key checks in transforms

[PierreQuinton]

• Add method check_and_get_keys to all transforms
• Move checks related to keys from init to check_and_get_keys, and make them recursive
• Remove required_keys and output_keys properties
• Remove check_keys_are call on the input keys in call and remove check_keys_are from TensorDict
• Replace _compute with call
• Change some tests of init into equivalent tests of check_and_get_keys
• Add new tests for check_and_get_keys
• Extract the creation of the transform in backward and mlt_backward into their _create_transform functions
• Rename _make_task_transform to _create_task_transform for uniformity
• Add test_check_create_transform for these functions
• Add changelog entry

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files with missing lines	Coverage Δ
src/torchjd/autojac/_transform/_differentiate.py	100.00% <100.00%> (ø)
src/torchjd/autojac/_transform/accumulate.py	100.00% <100.00%> (ø)
src/torchjd/autojac/_transform/aggregate.py	100.00% <100.00%> (ø)
src/torchjd/autojac/_transform/base.py	100.00% <100.00%> (ø)
src/torchjd/autojac/_transform/diagonalize.py	100.00% <100.00%> (ø)
src/torchjd/autojac/_transform/init.py	100.00% <100.00%> (ø)
src/torchjd/autojac/_transform/select.py	100.00% <100.00%> (ø)
src/torchjd/autojac/_transform/stack.py	100.00% <100.00%> (ø)
src/torchjd/autojac/_transform/tensor_dict.py	100.00% <ø> (ø)
src/torchjd/autojac/backward.py	100.00% <100.00%> (ø)
... and 1 more

🚀 New features to boost your workflow:

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

[ValerianRey]

LGTM.

In future PRs, we should:

• Do the same thing for runtime checks of TensorDict: move them from the init to a check method to make them optional.
• Make typing more strict in Transforms. If instead of having Iterable[Tensor] arguments we have for instance list[Tensor], we can safely defer the call to set(...) in check_and_get_keys, which will save some computations.

Note that the check that we removed from TensorDict (test_keys_are) is actually not necessary to replace by extra unit tests: we test all transforms well enough to ensure that their output keys are what we expect. No need for another PR for this.