arrow

GH-41361: [C++][Parquet] Optimize DelimitRecords by batch execution when max_rep_level > 1 (#41362)



### Rationale for this change

We uses Parquet to store nested types. When doing benchmarks, the nested types spend half time on `DelimitRecords`. The flamegraph can be seen in the issue. It can be reproduced when running `parquet-column-reader-benchmark`

This patch optimize DelimitRecords by batch execution. The previous code is slow because of branching. This patch changes branch to batch execution.

This could be a bit slower on some scenerio, but generally it makes DelimitRecords faster.

### What changes are included in this PR?

Change logic of DelimitRecords to batch execution.

### Are these changes tested?

Already has tests

### Are there any user-facing changes?

Yes

* GitHub Issue: #41361

Lead-authored-by: mwish <maplewish117@gmail.com>
Co-authored-by: Antoine Pitrou <pitrou@free.fr>
Signed-off-by: Antoine Pitrou <antoine@python.org>