remove extra copy of log-likelihood in NumPyroConverter

[andrewdipper]

Removes extra copy to reduce memory requirements for large models/datasets. If the copy is desirable for a reason I missed feel free to delete this. Let me know if any changes are needed

As a side note the default automatic numpyro log-likelihood calculation was a bit of a surprise (as far as unexpected memory explosions go).

• Follows official PR format
• Code style correct (follows pylint and black guidelines)

---

📚 Documentation preview 📚: https://arviz--2347.org.readthedocs.build/en/2347/

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.84%. Comparing base (cfbfbeb) to head (bb33953).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2347   +/-   ##
=======================================
  Coverage   86.84%   86.84%           
=======================================
  Files         123      123           
  Lines       12745    12745           
=======================================
  Hits        11068    11068           
  Misses       1677     1677           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[andrewdipper]

Also I realize the checks don't pass but it looks like some sort of conda environment failure. I'm not sure what to do about that, thanks

[OriolAbril]

It looks like you want to change the library default too: #2155. In the meantime you can create an rcfile so False is the default for you, https://oriolabrilpla.cat/en/blog/posts/2020/rcParams.html is a bit old but it should help, rcParams haven't changed much.

[OriolAbril]

LGTM, would be great to have some input from someone more familiar with numpyro. cc @fehiepsi maybe?

[fehiepsi]

Thanks, @andrewdipper!