Enzyme and Mooncake fwd mode support

[kshyatt]

Added forward rules and tests. Might be nice for doing forward mode AD for TNR in the future.

[codecov]

Codecov Report

❌ Patch coverage is 53.21101% with 51 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...orOperationsEnzymeExt/TensorOperationsEnzymeExt.jl	0.00%	51 Missing ⚠️

Files with missing lines	Coverage Δ
...erationsMooncakeExt/TensorOperationsMooncakeExt.jl	96.63% <100.00%> (+3.19%)	⬆️
...orOperationsEnzymeExt/TensorOperationsEnzymeExt.jl	0.00% <0.00%> (ø)

🚀 New features to boost your workflow:

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

[Jutho]

Looks good to me. I think these rules can be relaxed to not be restricted to AbstractArray subtypes; they would then immediately work for TensorMap as well.

This then also makes me trying to recall why we restricted the reverse rules? The only "array"-specific operation in most of these rules seems to be Mooncake.arrify, but since TensorMaps support this, would it not simplify our codebase to also relax the reverse rules?

[Jutho]

Also, these tests are currently not run. TensorOperations doesn't use ParallelTestRunners and the automatic detection stuff yet, so you have to manually include the files in runtests.jl

[kshyatt]

Wait, aren't all the tests in the test/mooncake.jl and test/enzyme.jl test files which are run?

[lkdvos]

I think we had to redo them for the tensormaps because of the additional fermionic signs...

[kshyatt]

I think the GitHub actions aren't uploading coverage data properly, I opened another PR about it

[Jutho]

Wait, aren't all the tests in the test/mooncake.jl and test/enzyme.jl test files which are run?

Oh yes, I thought these were new files, and only saw the zero% PR coverage message, so I inferred that the tests did not run.

[Jutho]

I think we had to redo them for the tensormaps because of the additional fermionic signs...

Damn you, fermions.

[kshyatt]

I've rebased on top of the Codecov PR, let's see if this looks a little better. Unfortunately the Enzyme code will still be uncovered...

[kshyatt]

There we go, looks like the Codecov is working again

[kshyatt]

@Jutho does this look good to merge ?

[Jutho]

Somehow it still seems as if none of the Enzyme changes are covered, whereas 100% of the Mooncake changes are.

[kshyatt]

Yes, unfortunately due to weird lowering behaviour, the Julia coverage cannot "see" into the Enzyme rules. We see the same behaviour over at MAK.