Reproducer for a bug in parquet 57.1.0+ where RowSelectionStrategy::Mask attempts to read skipped pages.

cargo test

Parquet error: Invalid offset in sparse column chunk data: 754, no matching page found.
If you are using a `SelectionStrategyPolicy::Mask`, ensure that the OffsetIndex is provided when creating the InMemoryRowGroup.

Test data is located in test.parquet.

See create_parquet binary for code that was used to create it.

• Columns: time (timestamp), tag (string)
• 2 row groups, 300 rows each

Tag column:

• 3 tag values 'a', 'b', 'c' with rows sorted by tag values
• When querying tag IN ('a', 'c') this creates selector list: [select 100, skip 100, select 100]

Time column:

• Times interleaved: even rows in predicate range, odd rows out of predicate range
• When querying on a time range this creates a sparse selection to trigger Mask strategy

Note: The page size is set so that the skip section (tag='b') contains at least one full page.

The bug occurs when:

1. A predicate uses RowSelectionStrategy::Selectors with a RowSelector list that skips an entire page.
2. Another predicate uses RowSelectionStrategy::Mask by triggering the mask run-length threshold.
3. The column with RowSelectionStrategy::Mask is not in the output projection, so should_force_selectors will not force it to use RowSelectionStrategy::Selectors.
4. The mask strategy will try to fetch pages that were skipped, resulting in an error.

• apache/arrow-rs#8733
• apache/arrow-rs#8845