Fix PG-means

[anripa]

These changes are made to fix the logic or to improve it:

• select the largest log likelihood model when updating (instead of lowest)
• use log likelihood for comparing models instead of the lower bound (following the original paper)
• improve new center proposal by picking the lowest log likelihood points (the likelihood is better function to test the goodness of fit than min max probability)
• fix possible bias in random projections by using randn instead of rand
• scale critical value if downsampling data

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.85%. Comparing base (4a36821) to head (90e8d67).
⚠️ Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #112      +/-   ##
==========================================
+ Coverage   92.80%   92.85%   +0.04%     
==========================================
  Files          63       63              
  Lines        9135     9232      +97     
==========================================
+ Hits         8478     8572      +94     
- Misses        657      660       +3     

Files with missing lines	Coverage Δ
clustpy/partition/pgmeans.py	100.00% <100.00%> (ø)

... and 26 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.