[ITensors] Call out to TensorOperations.jl for optimal sequence finding

[JoeyT1994]

Description

This PR removes the internal optimal_contraction_sequence(As) functions in ITensors.jl and outsources to the optimal tree function from TensorOperations.jl. Therefore optimal_contraction_sequence requires TensorOperations to be loaded, which is now a weak dependency of ITensors.jl.

Contraction sequence finding is now much faster, with orders of magnitude speedups manifesting for even relatively small vectors of Tensors.

How Has This Been Tested?

The tests from test/lib//ContractionSequenceOptimization/test_itensor_contract.jl have moved to test/ext/ITensorsTensorOperationsExt/runtests.jl and reflect the changes made to the optimal_contraction_sequence function.

[mtfishman]

Can you update the docs (https://github.com/ITensor/ITensors.jl/blob/main/docs/src/ContractionSequenceOptimization.md) as well?

[JoeyT1994]

I updated the docs. I also updated the version to 0.8.10 (not sure if this constitutes a sufficiently breaking change to change to 0.9)

[mtfishman]

I updated the docs. I also updated the version to 0.8.10 (not sure if this constitutes a sufficiently breaking change to change to 0.9)

I think we should update to v0.9, since it is technically a breaking change. Could you add a NEWS item about that here: https://github.com/ITensor/ITensors.jl/blob/main/README.md#news?

[JoeyT1994]

Sure, I went to 0.9.0 and added a news item. Not sure why the documentation check is failing though...

[mtfishman]

Sure, I went to 0.9.0 and added a news item. Not sure why the documentation check is failing though...

It's a bit annoying, but when we make breaking releases we also have to bump that compat entry in docs/Project.toml:

ITensors.jl/docs/Project.toml

Line 11 in c10e4e4

ITensors = "0.8.4"

.

[mtfishman]

Huh not sure what's going on with the docs... Maybe it is better to just remove the compat entry for ITensors from docs/Project.toml for now and we can look into that later.

[mtfishman]

Hmm, the docs build is erroring saying it can't find the docstring for ITensors.optimal_contraction_sequence, not sure what's going on there.

[mtfishman]

I see, maybe the issue is that the docstring is defined in the package extension, probably it should be defined on the version of the function in src/tensor_operations/tensor_algebra.jl.

[mtfishman]

Looks good, thanks @JoeyT1994!