[Remenby31]

Closes #4312

Rationale for this change

data_file_statistics_from_parquet_metadata built a column's bounds from the row groups that report a min/max and silently skipped a row group whose statistics have no min/max. PyArrow writes such statistics when a value is larger than 4 KB. The stored bounds then exclude that row group's values, and the metrics evaluator prunes the file from scans that should match it.

With this change, a row group that holds non-null values but reports no min/max drops the column's bounds for the file. The column keeps its null and value counts, because the row group still reports them. A row group whose values are all null still has no effect on the bounds.

Are these changes tested?

Yes. test_bounds_absent_when_a_row_group_with_values_has_no_min_max writes two row groups per column. One column has a 5,000-character value in its second row group, and the other column has an all-null first row group. The test fails without the change: the first column keeps the bound b'a'. tests/io/test_pyarrow_stats.py and the unit tests of tests/io/test_pyarrow.py pass, and prek passes on the changed files.

Are there any user-facing changes?

Data files written from now on have no bounds for such a column, so scans no longer skip their matching rows. Files already written keep their bounds.