Fix recognition of tuning stat in .to_inferencedata() method

[michaelosthege]

This fixes two bugs, one of which only got triggered by the first.

Since PyMC v5.7.0 the "tune" stat is sampler-wise, and named "sampler_0__tune" or similar.

However, the .to_inferencedata() implementation here was not updated, and kept looking for "tune".
When this is not available, an all-false mask was supposed to be used, but since np.repeat was used, this was just an empty array, resulting in all draws getting sliced away all the time.

The change here is two-fold:

• 🐛 Use np.full to create the mask if no tune stat was found.
• 🐛 Look for stats named "tune" or ending with "__tune" to be compatible with PyMC before and after 5.7.0.
• 🔧 Added an __all__ to hide non-public parts of the .core API.
• 🆙 Version incremented to 0.5.2

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.45%. Comparing base (9fa6fb6) to head (3487ed7).
Report is 13 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #103      +/-   ##
==========================================
+ Coverage   95.80%   96.45%   +0.64%     
==========================================
  Files           8        8              
  Lines         501      508       +7     
==========================================
+ Hits          480      490      +10     
+ Misses         21       18       -3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.