Rename logprob/joint_logp to logprob/basic and move logcdf and icdf functions there

[ricardoV94]

This PR is a spinoff of #6597, to facilitate review.

Important changes:

• logprob/joint_logprob module was renamed to logprob/basic.py
  It holds all the user-facing probability constructors
• The dispatch call helpers in abstract are now private-named functions and their behavior is clarified
• logp no longer calls the heavier factorized_joint_logprob. It immediately dispatches on the IR

[codecov]

Codecov Report

Merging #6599 (7821ebe) into main (2fcce43) will decrease coverage by 0.02%.
The diff coverage is 97.05%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6599      +/-   ##
==========================================
- Coverage   92.03%   92.02%   -0.02%     
==========================================
  Files          93       93              
  Lines       15746    15750       +4     
==========================================
+ Hits        14492    14494       +2     
- Misses       1254     1256       +2     

Impacted Files	Coverage Δ
pymc/distributions/mixture.py	95.40% <66.66%> (-0.03%)	⬇️
pymc/logprob/abstract.py	98.80% <94.73%> (+1.21%)	⬆️
pymc/distributions/bound.py	100.00% <100.00%> (ø)
pymc/distributions/continuous.py	97.70% <100.00%> (ø)
pymc/distributions/discrete.py	98.73% <100.00%> (ø)
pymc/distributions/timeseries.py	94.54% <100.00%> (ø)
pymc/distributions/truncated.py	99.30% <100.00%> (+<0.01%)	⬆️
pymc/logprob/__init__.py	100.00% <100.00%> (ø)
pymc/logprob/basic.py	98.92% <100.00%> (ø)
pymc/logprob/cumsum.py	100.00% <100.00%> (ø)
... and 7 more

... and 1 file with indirect coverage changes

[Armavica]

Let me see if I understand this correctly:

We enter this NonImplementedError in cases where the rv.owner.op has a type that was not registered anywhere with @_logprob.register(...), and so that falls back to the default (non)-implementation in abstract.py.

In these cases, why would the FunctionGraph approach work?

[ricardoV94]

It falls back to the intermediate representation (IR) which converts stuff like Exp(Beta) into MeasurableExp(Beta) (or whatever is deemed valid), which itself has a logp method dispatched.

That's basically how inference for non basic RVs is carried out.

[ricardoV94]

Ready for re-review

[Armavica]

Ok, I think I understand this well enough. I like the clearer hierarchy of functions and the reduction of user-facing logp-related functions!