Test QR rules with CUDA

[kshyatt]

No description provided.

[kshyatt]

OK, this is now working for QR on full CuMatrix, but there are still scalar indexing errors for Diagonal because dispatch for SubArray{T, 2, Diagonal{T, CuVector{T}} is awful. I'd be ok with merging this as-is if the pullback modifications look ok, and then come back for another swing at the Diagonal case.

[lkdvos]

Would it make sense to introduce a project_triu(!) and/or norm_triu function for this, and just overload the GPU ones to allocate?

As a side note, we might at some point try and not generate any of the gauge dependency checks if the tolerance is set to 0, as a way to disable the warnings, or alternatively put the entire thing in a @warn block so we can disable that, since these checks are starting to feel a bit expensive if we need to start allocating.

[kshyatt]

Would it make sense to introduce a project_triu(!) and/or norm_triu function for this, and just overload the GPU ones to allocate?

Yeah I think that could be a nice way to handle it, however I'm unsure if it's that helpful if it's only used in these checks.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/pullbacks/qr.jl	97.02% <100.00%> (+1.47%)	⬆️

🚀 New features to boost your workflow:

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

[kshyatt]

The Enzyme windows fail is happening on main and seems unrelated