fix bugged TV denoising tests

[tjlane]

Some unit tests targeting the TV denoising code were broken or flakey.

The idea behind these tests is to take a simple "ground truth" map, add some Gaussian noise, and then run TV denoising. Ideally, the output should be closer to the noise-free "ground truth" than the input.

But these tests were not faithfully reporting this because the map comparison was done by computing a normalized RMS between the two maps. The Pearson correlation, computed in real space, appears to provide a more robust signal, and this PR implements this change.

During the implementation, I realized that I was confused by the rsmap.Map methods. There is a .to_numpy() method inherited from pandas that provides the F, PHI, SIGF columns as numpy arrays. Useful, but for a moment, I thought I was getting real space maps. Therefore, this PR additionally implements a handy function that computes a 3d_numpy_map, which hopefully will make this easier and clearer going forward.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.63%. Comparing base (6f0ab62) to head (343d73d).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #100   +/-   ##
=======================================
  Coverage   95.63%   95.63%           
=======================================
  Files          15       15           
  Lines         940      940           
=======================================
  Hits          899      899           
  Misses         41       41           

Flag	Coverage Δ
unittests	95.63% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

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