Improve Numerical Stability

[kmdalton]

Several users have been reporting numerical issues with careless. Recently, I was able to reproduce them, and I tracked the issue down to an overflow in the scale parameter of the normal distribution which parameterizes the reflection scales. The default bijector here has been the exponential for sometime. However, apparently it does not work stably for some data sets (probably those with particularly bright reflections). One solution would be to normalize the intensities in the likelihood calculation. This might be worth exploring in the future. For now, I have changed the default bijector back to softplus and added a command line flag --scale-bijector={exp,softplus}.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 91.66667% with 1 line in your changes missing coverage. Please review.

Project coverage is 80.80%. Comparing base (0d5e4a4) to head (9ea412a).
Report is 21 commits behind head on main.

Files with missing lines	Patch %	Lines
careless/io/manager.py	87.50%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #167      +/-   ##
==========================================
+ Coverage   78.69%   80.80%   +2.10%     
==========================================
  Files          52       52              
  Lines        2347     2386      +39     
==========================================
+ Hits         1847     1928      +81     
+ Misses        500      458      -42     

Flag	Coverage Δ
unittests	80.80% <91.66%> (+2.10%)	⬆️

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.