Remove workaround in InferenceData conversion; require ArviZ >=0.11.4

[michaelosthege]

This PR removes a workaround in the InferenceData conversion and at the same time fixes a bug that was present in the workaround.

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? → possibly ArviZ API changes
• important background, or details about the implementation →
• are the changes—especially new features—covered by tests and docstrings? → yup
• linting/style checks have been run → aye

[OriolAbril]

doesn't need to be in this PR, but we can remove this argument. densitydist api will be fixed in v4 so this is no longer necessary.

[codecov]

Codecov Report

Merging #5060 (b18a09e) into main (33dd132) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main    #5060      +/-   ##
==========================================
- Coverage   78.42%   78.41%   -0.01%     
==========================================
  Files         130      130              
  Lines       24475    24461      -14     
==========================================
- Hits        19195    19182      -13     
+ Misses       5280     5279       -1     

Impacted Files	Coverage Δ
pymc/backends/arviz.py	88.75% <100.00%> (-0.60%)	⬇️
pymc/tests/test_distributions.py	96.34% <100.00%> (+0.07%)	⬆️

[michaelosthege]

thanks for dealing with the dependencies, @OriolAbril !

I just rebased on main to resolve conflicts. Also I rearranged the commits such that the dependency upgrade is now all in one. Let's hope the CI likes it.

[michaelosthege]

@OriolAbril can you check the diff again? I had to mark a test as XFAIL because I couldn't reproduce the float64/float32 casting error locally.
My feeling is that this is not critical, and because it's a good thing to have less strict dependencies we can go ahead?

[twiecki]

Can merge once we get green light.

[twiecki]

👍