Return InferenceData when there are no variables sampled and extend=True

[OriolAbril]

Description

When there are no variables to sample and extend_inferencedata=True the returned object
is trace at that point in the code which in the common case of providing an InferenceData
to sample_posterior_predictive is only its posterior group. Consequently:

with model:
    idata = pm.sample_posterior_predictive(idata, var_names=[], extend_inferencedata=True)
# now `idata` is the xarray Dataset corresponding to the posterior group

This PR changes the behaviour so idata above continues to be the provided idata and adds
a test for it.

Related Issue

I considered it was easier to open the PR directly instead of first issue then PR.

Checklist

• Checked that the pre-commit linting/style checks pass
• Included tests that prove the fix is effective or that the new feature works
• If you are a pro: each commit corresponds to a relevant logical change

Type of change

• New feature / enhancement
• Bug fix
• Documentation
• Maintenance
• Other (please specify):

---

📚 Documentation preview 📚: https://pymc--7509.org.readthedocs.build/en/7509/

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.43%. Comparing base (4300be1) to head (e216e02).
Report is 98 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #7509   +/-   ##
=======================================
  Coverage   92.43%   92.43%           
=======================================
  Files         103      103           
  Lines       17109    17109           
=======================================
+ Hits        15814    15815    +1     
+ Misses       1295     1294    -1     

Files with missing lines	Coverage Δ
pymc/sampling/forward.py	96.15% <100.00%> (+0.42%)	⬆️