refactor: Use matrix-matrix interface in LAPACK solve

[dkachuma]

This PR replaces multiple matrix-vector solves with a single matrix-matrix solve following PR3173. This also uses a version of the LAPACK call the destroys the coefficient matrix and rhs matrix to avoid further allocation in BlasLapackLA.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 56.01%. Comparing base (ececf78) to head (bdf0c0d).
Report is 92 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #3234      +/-   ##
===========================================
- Coverage    56.02%   56.01%   -0.01%     
===========================================
  Files         1053     1053              
  Lines        89123    89115       -8     
===========================================
- Hits         49927    49917      -10     
- Misses       39196    39198       +2     

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

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[CusiniM]

I personally preferred having the rhs and the solution into two separate arrays. I am guessing that this is the lapack interface though... We could keep the separation on the GEOS side though.

[dkachuma]

The separate rhs/solution interface on BlasLapackLA still exists. This change destroys the rhs but avoids having to allocate extra space for solution and rhs knowing that we are not going to need the rhs or the matrix after the solve.

[CusiniM]

is the allocation the same? A single array2d instead of 2 array1d?

[dkachuma]

My mistake. The allocation was on the matrix. With the previous interface, everytime we call the solve, additional space had to be allocated for the LU factors of the matrix. This way there is no extra space allocated since it's a single solve and the space allocated for the matrix is used for the LU factors (destroying the matrix but once we have solved we don't care anymore).

With the single column by column solve it would have been nice to factorize the matrix and then use the LU factors on each column but we don't have that interface. Instead I added an interface where you can do an in-place matrix-matrix solve.