Add a type guard for intX (#4569)

[michaelosthege]

These are the changes from #4569 cherry-picked from 3447619 on the v3 branch.

Closes #4279

Checklist

• Explain important implementation details (see Add a type guard for intX #4569)
• 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

• None

Bugfixes / New features

• Integer dtypes are now not downcasted in intX

Docs / Maintenance

• None

[codecov]

Codecov Report

Merging #6319 (7c86ed7) into main (c4a79ef) will decrease coverage by 11.83%.
The diff coverage is 100.00%.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #6319       +/-   ##
===========================================
- Coverage   94.24%   82.41%   -11.84%     
===========================================
  Files         111      111               
  Lines       23930    23936        +6     
===========================================
- Hits        22553    19727     -2826     
- Misses       1377     4209     +2832     

Impacted Files	Coverage Δ
pymc/aesaraf.py	87.34% <100.00%> (-7.28%)	⬇️
pymc/distributions/discrete.py	99.23% <100.00%> (+<0.01%)	⬆️
pymc/tests/distributions/util.py	90.86% <100.00%> (ø)
pymc/tests/backends/test_base.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/backends/test_ndarray.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/distributions/test_logprob.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/distributions/test_shape_utils.py	0.00% <0.00%> (-99.77%)	⬇️
pymc/tests/distributions/test_timeseries.py	0.00% <0.00%> (-99.52%)	⬇️
pymc/tests/distributions/test_mixture.py	0.00% <0.00%> (-99.36%)	⬇️
pymc/tests/step_methods/hmc/test_quadpotential.py	0.00% <0.00%> (-95.82%)	⬇️
... and 15 more

[ricardoV94]

The original issue was about not downcasting int64 to int32, I think it's fine to still apply the old rule in we don't have either of those

[michaelosthege]

The original issue was about not downcasting int64 to int32, I think it's fine to still apply the old rule in we don't have either of those

I'm not sure what you mean. Should we proceed here, or close this and #4279 was wontfix?

[ferrine]

This issue was raised because of some discrete likelihoods were overflowed in float32 mode essentially due to int downcasting. The typeguard prevented unintentional downcast if input is already int

[ricardoV94]

Basically we want to prevent downcasting but not upcasting which the original fix also prevented. If you still allow upcasting you won't see the issues you were trying to fix in the new commit

[michaelosthege]

stale