fix: Reading Parquet with Null dictionary page

[coastalwhite]

This fixes an issue with some Parquet writers that write dictionary pages for Null arrays (why?? I have no idea?).

Fixes #18085.
Fixes #18079.

Possibly also #18061.

[codecov]

Codecov Report

Attention: Patch coverage is 57.14286% with 3 lines in your changes missing coverage. Please review.

Project coverage is 80.33%. Comparing base (3dda47e) to head (de3db83).
Report is 10 commits behind head on main.

Files	Patch %	Lines
...tes/polars-parquet/src/parquet/read/compression.rs	50.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #18112      +/-   ##
==========================================
- Coverage   80.37%   80.33%   -0.04%     
==========================================
  Files        1496     1496              
  Lines      197542   197683     +141     
  Branches     2820     2821       +1     
==========================================
+ Hits       158771   158811      +40     
- Misses      38249    38351     +102     
+ Partials      522      521       -1     

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

[ritchie46]

Nice.

Can you add this test?

import pandas as pd
test = pd.DataFrame([{'A':np.NaN,'B':3,'C':None},{'A':np.NaN,'B':None,'C':None}])

test.to_parquet('~/mre.parquet')

pl.scan_parquet('~/mre.parquet').collect()  

Slightly adapted so that it writes to a bytes buffer.

[coastalwhite]

Did it 👍