Feature Statistics: Fix sorting

[janezd]

Issue

Fixes #5329. At least I think so. After tinkering with a number of glitches, I'm no longer sure what fixed it.

Description of changes

1. Put categorical variables at the bottom, because they are less interesting for this widget. But in particular, put them at the bottom when sorting by min and max, because these values don't apply to them.
2. Sorting by variable names didn't work because the result of argsort can be used for indexing, not sorting. However, calling argsort twice computes the inverse permutation.
3. Ensure stable order also when sorting by single number in descending order. The code used to invert the stable order. :)
4. Use ascending, not descending order by default.

These changes, at least the last one, breaks the tests. I'm gonna fix them when we confirm the changes.

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

Merging #5331 (251acaf) into master (69e580d) will decrease coverage by 0.01%.
The diff coverage is 63.63%.

@@            Coverage Diff             @@
##           master    #5331      +/-   ##
==========================================
- Coverage   85.35%   85.34%   -0.02%     
==========================================
  Files         301      301              
  Lines       62035    62039       +4     
==========================================
- Hits        52950    52946       -4     
- Misses       9085     9093       +8