Avoid redundant full-cube-passes when region submasking

[keflavich]

Region sub-masking has been using minimal_subcube, which shouldn't be necessary b/c we already downselect to the minimal bounding box.

Also, it was creating 3D masks when only 2D is needed - the broadcasting might therefore not have been done lazily.

[codecov-commenter]

Codecov Report

Merging #801 (122397f) into master (f92fe7d) will decrease coverage by 0.03%.
The diff coverage is 16.66%.

❗ Current head 122397f differs from pull request most recent head 8664b89. Consider uploading reports for the commit 8664b89 to get more accurate results

@@            Coverage Diff             @@
##           master     #801      +/-   ##
==========================================
- Coverage   77.94%   77.90%   -0.04%     
==========================================
  Files          24       24              
  Lines        5853     5856       +3     
==========================================
  Hits         4562     4562              
- Misses       1291     1294       +3     

Impacted Files	Coverage Δ
spectral_cube/analysis_utilities.py	83.83% <ø> (ø)
spectral_cube/spectral_cube.py	76.59% <16.66%> (-0.16%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f92fe7d...8664b89. Read the comment docs.

[keflavich]

This produces WCS mismatches only on windows. That seems unlikely; either this is a problem with the test grid or a genuine problem with how WCS behaves on windows machines?

[keflavich]

Ready for final review

[e-koch]

Small docs change. Otherwise looks good!