Fix logcdf and icdf derivations for non-monotonically increasing transformations

[ricardoV94]

#6597 was only valid for strictly monotonically increasing transformations.

This PR fixes the derivations for monotonically decreasing transformations and adds a special case for logcdf of negative powers for distributions that allow negative values.

Also fixes the derivations for cosh transformed variables as that is not injective (CC @LukeLB)

Refactored a bit the test module with Test classes. It was growing a tad too unmanageable.
The first two commits are just refactoring and bugfixes come after

---

📚 Documentation preview 📚: https://pymc--6850.org.readthedocs.build/en/6850/

[codecov]

Codecov Report

Merging #6850 (bafe249) into main (ccad4c8) will decrease coverage by 0.02%.
Report is 2 commits behind head on main.
The diff coverage is 91.17%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6850      +/-   ##
==========================================
- Coverage   92.05%   92.03%   -0.02%     
==========================================
  Files          96       96              
  Lines       16369    16398      +29     
==========================================
+ Hits        15068    15092      +24     
- Misses       1301     1306       +5     

Files Changed	Coverage Δ
pymc/logprob/transforms.py	94.42% <90.90%> (-0.27%)	⬇️
pymc/testing.py	91.46% <100.00%> (ø)

... and 2 files with indirect coverage changes

[larryshamalama]

Apologies for the delay... Looks great @ricardoV94 :) No major questions from me this time. Admittedly, I didn't look too carefully at the tests, I have full confidence in them 😅