refactor(autojac): Stop cloning in Accumulate

[ValerianRey]

This follows our discussion on whether we should clone or not in Accumulate. The change alone made a test fail, but this test was precisely a very weird usage of Accumulate, where the same tensor dict is accumulated multiple times. So I also changed the test. I think this could have a benefit on memory usage, but not sure if this benefit can happen when memory is at its peak usage (at this point, the Jacobian should be aggregated already, so not in memory anymore, so maybe we don't care that much about an extra gradient in memory there until the completion of this function). But I remember correctly, we used to have some out of memory errors during the Accumulate call, so I might be wrong about this.

Anyway, it seems like a strict improvement.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/torchjd/autojac/_transform/_accumulate.py	100.00% <100.00%> (ø)

🚀 New features to boost your workflow:

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