fix(rust): Fix null handling in collect_statistics of parquet reader

[codesorcery]

Fixes #20361

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 78.90%. Comparing base (6e4d717) to head (c7808d3).
Report is 12 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #20362      +/-   ##
==========================================
- Coverage   79.13%   78.90%   -0.23%     
==========================================
  Files        1572     1567       -5     
  Lines      219839   219920      +81     
  Branches     2462     2465       +3     
==========================================
- Hits       173970   173537     -433     
- Misses      45301    45815     +514     
  Partials      568      568              

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

[ritchie46]

@coastalwhite is this ok?

[coastalwhite]

I will look at it. I would really like to make a test with an MRE before merging this. I am pretty sure if I look at the code for a second I can find one.

[coastalwhite]

This is an okay fix for now. It should really be handled in #20203 by filling in a null.

[coastalwhite]

Actually, I think this should just be fixed by #20203. This kind of makes it impossible to do predicate pushdown with missing columns.

[codesorcery]

@coastalwhite thanks for the quick review and the MRE!
However, I think there was a mistake in the test. You were filtering on pl.col.a == pl.col.b, which is true for the data in the first data frame and false for the data in the second data frame. However, the assertion expects the result to contain the data of both data frames.
I updated the test, so that in both data frames there are rows which the filter applies to.