Fix dependence of Uniform logp on bound method

[ricardoV94]

This PR fixes #4499

The Uniform and Discrete Uniform logp methods depended on the dist_math::bound method for correct evaluation of multiple values. This is problematic because the bound can be disabled by setting check_bounds = False in pm.Model().

Other changes:

1. I tried to improve the documentation of the bound method to emphasize this. I also tried to make it a bit more clear what the "broadcast_conditions" secret keyword argument means and when it should be used. I would suggest this to be made an optional keyword argument in the function call of bound, and perhaps even rename it to multivariate=False to make it more obvious when should it be used. Let me know if you agree and I can add it to this PR as well.

2. I tweaked the docstring of check_bounds to emphasize this should not be used for models containing discrete variables as some of these depend on bound checks when sampled (unless they have specific samplers that cannot propose invalid values or the logp happens to behave nicely when given out-of-bound points)

3. Removed rendudant tt.all() calls in the bound conditions of several multivariate distributions.

---

Do we need a release note?

Depending on what your PR does, here are a few things you might want to address in the description:

• what are the (breaking) changes that this PR makes? None
• important background, or details about the implementation None
• are the changes—especially new features—covered by tests and docstrings? Tests cover known bugs. Docstrings improved
• linting/style checks have been run
• consider adding/updating relevant example notebooks
• right before it's ready to merge, mention the PR in the RELEASE-NOTES.md

[twiecki]

LGTM - thanks! Still needs a line in the release notes.

[michaelosthege]

I think we shouldn't include the entire section: Some changes in v3 (the stats and plotting functions) are not included on the master branch. See #4528.

Instead, we could just mark the "also-in-v3-changes" with "(released in 3.11.2)"?

[michaelosthege]

@MarcoGorelli do you know of best practices when it comes to changelog files in a world with multiple branches and backporting?

[ricardoV94]

I don't know. What about just repeating again this line from V3.11.0?

• ⚠ Many plotting and diagnostic functions that were just aliasing ArviZ functions were removed (see 4397). This includes pm.summary, pm.traceplot, pm.ess and many more!

[ricardoV94]

That's the only bump between V3.11.1 -> V3.11.2 -> V4.0.0 no?

[michaelosthege]

For now we could do that. I'll make sure to do the PR such that pm.plots and pm.stats are re-introduced on master too.
Then we can edit the release notes again..