Adjust test to avoid inconsistent results on windows for test_scale_and_filter....

[jbeilstenedmands]

Fixes #1895

From the logs, the difference between Windows/Mac appears to start in the second round of scaling, in the first run of LBFGS minimisation. After around 30 steps, the RMSDs start changing in the 4th decimal place, so a slightly different solution is found, which causes the delta-cc-half scores of the two worst groups to change order and therefore the resultant dataset after filtering contains different sets of images, leading to large changes in Rpim etc.
Although this dataset is rather thin and radiation damaged, so the minimisation might be expected to be less stable, I have kept this as the test case and adjusted the test parameters to avoid relying on the behaviour past the point of divergence, while still testing the same overall functionality.

[ndevenish]

I've tested with the prebuilt-cctbx/dxtbx cmake build on windows, and this is indeed now passing. Thank you.

[ndevenish]

And on mac-arm!

[codecov]

Codecov Report

Merging #2104 (c3643e8) into main (9248809) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main    #2104      +/-   ##
==========================================
- Coverage   80.31%   80.31%   -0.01%     
==========================================
  Files         576      576              
  Lines       65245    65241       -4     
  Branches     9208     9208              
==========================================
- Hits        52404    52396       -8     
- Misses      10804    10805       +1     
- Partials     2037     2040       +3