Sparse matrix testing and added features

[ShreyasKhandekar]

This PR adds more features to the SparseMatrix Module:

1. fill_vals : Populates the sparse matrix with a 1D array which has an equal number of elements as the number of non zero values in the sparse matrix. The order in which the values are assigned is the natural iteration order of the sparse matrix (CSC vs CSR)
2. to_pdarray: Converts a sparse matrix to 3 pdarrays: row indices, col indices, and the values at the indices. The size of each array is equal and is equal to the nnz in the sparse array.

This PR also adds a few simple tests for the above utilities, as well as for sparse_matrix_matrix_mult

We also add a new compatibility shim ArkoudaSparseMatrixCompat.chpl, to make this work with 2.0. Note that even though this works with 2.0, performance will be worse than later releases. This is because chapel-lang/chapel#24391 and chapel-lang/chapel#25243 were two PRs that improved the performance of sparse computations which are not a part of 2.0

We also change the SparseMatrix module to be compiled by default in Arkouda.

Tested single and multilocale

[ShreyasKhandekar]

I see the chapel portability tests failing, given that a lot of the new work done by Brad in order to support these sparse matrix computations in Arkouda came after 2.0, I'm not sure what we can do to resolve this because it's not compatible AFAIK. I may be mistaken here and @bradcray might have a better idea.

[bmcdonald3]

Looks good assuming you've tested in multilocale, thanks Shreyas

[ShreyasKhandekar]

Yup, testing passes with multilocale. Although if we scale it up to more than 8 locales, it will run out of memory due to the inefficient implementations used for now.

Also, I think there is something up with the CI:

tests/alignment_test.py ...FFF.....                                      [  4%]
tests/bigint_agg_test.py F....                                           [  4%]
tests/bitops_test.py FFFFFFFF..                                          [  5%]
tests/categorical_test.py ...........................                    [  6%]
tests/client_dtypes_test.py ..........F                                  [  7%]
tests/client_test.py ...........                                         [  7%]

It's been stuck here for over 1.5 hours, and many tests are failing that I didn't modify in any way.

Could we restart it?

[bmcdonald3]

Although if we scale it up to more than 8 locales, it will run out of memory

Is this on your laptop or in all environments? If we run out of memory at 8 locales, this will fail in nightly testing, which does 16 nodes on an XC

[ShreyasKhandekar]

This was on ChapDL. I thought only the benchmarks did 16 locale runs, not the correctness tests?

[bmcdonald3]

No, the correctness tests run at 16 locales as well

[ShreyasKhandekar]

Ah, okay. I will try to reduce the problem size or the memory requirement (or both) to see if I can make it work on 16 locales then. Thanks for the catch!

[ShreyasKhandekar]

Okay I think I've convinced myself that my previous tests where I was running out of memory and getting a segfault were because somehow the wrong (non-distributed) version of the sparse matrix matrix multiplication overload was being chosen. I've commented out that version so it doesn't get accidentally chosen. I've reduced the problem size anyway, just to be safe.

If the CI passes, I think we should be good to merge this.

[jeremiah-corrado]

Looks good, glad to see this going in!

Overall, I think it would be nice to get the server-side code converted over to the new function-registration framework at some point, but I definitely don't think that needs to happen for this PR.

Other than that, I just had a few minor notes:

[stress-tess]

looks good to me! thanks sheryas!