Add sparse matrix option with Compressed Sparse Column ordering

[mattldawson]

Adds Compressed Sparse Column ordering option for standard and vector sparse matrix ordering.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 98.34711% with 6 lines in your changes missing coverage. Please review.

Project coverage is 93.68%. Comparing base (cdfd2e7) to head (bda6189).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
...matrix_standard_ordering_compressed_sparse_row.hpp	96.29%	2 Missing ⚠️
...e_matrix_vector_ordering_compressed_sparse_row.hpp	96.72%	2 Missing ⚠️
...rix_standard_ordering_compressed_sparse_column.hpp	98.33%	1 Missing ⚠️
...atrix_vector_ordering_compressed_sparse_column.hpp	98.55%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #693      +/-   ##
==========================================
+ Coverage   93.54%   93.68%   +0.13%     
==========================================
  Files          55       57       +2     
  Lines        3796     3941     +145     
==========================================
+ Hits         3551     3692     +141     
- Misses        245      249       +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[sjsprecious]

Thanks @mattldawson for working on this new feature. Looks good to me.

I could also confirm that the GPU tests pass on Derecho's A100 GPU with that latest nvhpc/24.11 compiler.

[K20shores]

Why do we need this policy for the matrices?

[mattldawson]

One of the things we wanted to try was using CSC for the L matrix and CSR for the U matrix with the Doolittle algorithm. This let us try that out, but it didn't have a big impact (is this right @sjsprecious ?). But, we're not sure why yet, so I think we wanted to leave the option until we could better understand the performance results.

[sjsprecious]

Matt's right as I originally thought different policies for L/U matrices would make the performance better, but it was actually not. I still think after some optimizations we should see a benefit from this change but I could be wrong. As Matt said, let's leave this option here until we have a better understanding of the performance results.