Simplify Potential docstrings and examples

[ricardoV94]

This PR simplifies the examples and explanations used in the Potential docstrings

---

📚 Documentation preview 📚: https://pymc--6772.org.readthedocs.build/en/6772/

[codecov]

Codecov Report

Merging #6772 (515f7ab) into main (154f5b0) will decrease coverage by 0.46%.
The diff coverage is 100.00%.

❗ Current head 515f7ab differs from pull request most recent head a71fa6a. Consider uploading reports for the commit a71fa6a to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6772      +/-   ##
==========================================
- Coverage   91.89%   91.44%   -0.46%     
==========================================
  Files          95       95              
  Lines       16147    16147              
==========================================
- Hits        14839    14765      -74     
- Misses       1308     1382      +74     

Impacted Files	Coverage Δ
pymc/model.py	89.66% <100.00%> (-0.25%)	⬇️

... and 4 files with indirect coverage changes

[daniel-saunders-phil]

Hey Ricardo, is the first example is supposed to be executable?

data = stats.norm(0.05,1).rvs(50)

with pm.Model() as m8:
    x = pm.Normal("x", mu=0, sigma=1)
    y = pm.Normal("y", mu=x, sigma=1, observed=data)
    constraint = x >= 0
    potential = pm.Potential("x_constraint", pm.math.log(pm.math.switch(constraint, 1, 0)))
    
    t8 = pm.sample()

I get AttributeError: 'float' object has no attribute 'type' . Full error trace back: https://gist.github.com/daniel-saunders-phil/b6656f51b0fae351274711dcfb2641f2 . The other examples under Potential work fine. Also, if I change the switch from a hard constraint to a very firm constraint (pm.math.switch(constraint, 1, 0.1)), the example runs (with poor sampler performance). So it looks like the problem is unique to adding log of 0 to the potential.

[ricardoV94]

I think this was fixed in the very last release of PyTensor (1 or 2 days old) and I was probably already on it.

Can you update and check again?

[ricardoV94]

By the way I think you're looking at the old example (not that it matters for the error your stumbled upon)

Also these switch Potential models are not expected to sample well, it's just for illustration. The discontinuities they introduce are pretty bad for NUTS

[daniel-saunders-phil]

I think this was fixed in the very last release of PyTensor (1 or 2 days old) and I was probably already on it.

Can you update and check again?

Ahh okay you're on top of it. Doesn't look the latest version is available through conda-forge at the moment. I can only get 2.12.1. The new examples are also much nicer to read btw