Parquet statistics stats_null_count is wrong for large json string colums

stats_null_count always None when one of the columns values is about 46977 characters length   

**Reporter**: [Rotem Levi](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=rotem.l)
#### Original Issue Attachments:
- [part-00000-7e5dcc44-d176-42ab-98b9-664cace5d433-c000.snappy.parquet](https://issues.apache.org/jira/secure/attachment/13058765/part-00000-7e5dcc44-d176-42ab-98b9-664cace5d433-c000.snappy.parquet)

_{**Note**: *This issue was originally created as [PARQUET-2306](https://issues.apache.org/jira/browse/PARQUET-2306). Please see the [migration documentation](https://issues.apache.org/jira/browse/PARQUET-2502) for further details.*}

Start by reproducing the reported stats_null_count behavior with the attached part-00000-7e5dcc44-d176-42ab-98b9-664cace5d433-c000.snappy.parquet file and a column containing a roughly 46,977-character JSON string. Trace how Parquet statistics are read for that column; done means stats_null_count is reported correctly instead of always being None.

Written by the indexing model from the issue text.