Don't mask inputs to pm.Data containers

[michaelosthege]

What is this PR about?
I ran into a problem where NaN values in my data got masked away when creating pm.Data containers.
This was quite unexpected.

We need the masking for observed, but that's being taken care of in other places already.

Checklist

• Explain important implementation details 👆
• Make sure that the pre-commit linting/style checks pass.
• Link relevant issues (preferably in nice commit messages)
• Are the changes covered by tests and docstrings?
• Fill out the short summary sections 👇

Major / Breaking Changes

• NaN values passed to pm.Data are no longer masked automatically.

Bugfixes / New features

• None

Docs / Maintenance

• None

[codecov]

Codecov Report

Merging #6189 (dd5a1af) into main (244c37d) will decrease coverage by 0.02%.
The diff coverage is 98.76%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6189      +/-   ##
==========================================
- Coverage   93.04%   93.01%   -0.03%     
==========================================
  Files          91      101      +10     
  Lines       20813    22092    +1279     
==========================================
+ Hits        19365    20549    +1184     
- Misses       1448     1543      +95     

Impacted Files	Coverage Δ
pymc/aesaraf.py	89.66% <ø> (-3.88%)	⬇️
pymc/model.py	88.10% <ø> (-0.14%)	⬇️
pymc/parallel_sampling.py	85.76% <ø> (-0.05%)	⬇️
pymc/tests/backends/test_ndarray.py	100.00% <ø> (ø)
pymc/tests/smc/test_smc.py	100.00% <ø> (ø)
pymc/sampling.py	82.63% <85.71%> (-0.02%)	⬇️
pymc/distributions/multivariate.py	92.24% <91.66%> (-0.07%)	⬇️
pymc/stats/convergence.py	93.15% <93.15%> (ø)
pymc/distributions/shape_utils.py	98.66% <97.14%> (-0.29%)	⬇️
pymc/tests/variational/test_inference.py	99.02% <99.02%> (ø)
... and 29 more

[michaelosthege]

@ricardoV94 and I discussed this offline and didn't find a good solution that keeps pm.Data inputs unmasked while not breaking imputation or model_to_graphviz at the same time.

Without a good solution, and keeping in mind that @ricardoV94 wants to refactor how we manage observed variables I think we can close this PR.