fix: Fix to_pandas() on empty enum Series did not preserve enum dictionary

[nameexhaustion]

print(pl.Series(dtype=pl.Enum(["A"])).to_pandas())

# Before
Series([], Name: , dtype: category
Categories (0, object): [])
# After
Series([], Name: , dtype: category
Categories (1, str): ['A'])

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.39%. Comparing base (83d6a72) to head (13ee338).
⚠️ Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #26610      +/-   ##
==========================================
+ Coverage   80.76%   81.39%   +0.63%     
==========================================
  Files        1795     1795              
  Lines      244990   244992       +2     
  Branches     3079     3079              
==========================================
+ Hits       197871   199423    +1552     
+ Misses      46333    44783    -1550     
  Partials      786      786              

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[orlp]

What is this cast? Is this a pyarrow cast? Is this not supposed to be fixed upstream? Seems rather hacky all this.

[nameexhaustion]

Yes, pyarrow cast, indeed very hacky.

Not sure this would be considered a bug in pyarrow; I don't think they promise to preserve the dictionary across operations - it's similar to our old categoricals in that every chunk has its own local mapping.

I've pushed an update where we instead just dispatch to DataFrame.to_pandas, which does this cast using our own cast kernels in polars-compute.