bench(parquet): Add benchmarks for mask-backed RowSelection::and_then

[haohuaijin]

Which issue does this PR close?

• Part of Optimize RowSelection::and_then for dense masks #10967.

Rationale for this change

Add focused benchmarks for mask-backed RowSelection::and_then. These benchmarks measure the input shapes targeted by #10968 and make its performance results reproducible.

What changes are included in this PR?

• Add pseudo-random and clustered mask generators.
• Sweep outer and inner selectivity around the proposed thresholds.
• Sweep input lengths on both sides of the 8,192-row threshold.
• Cover non-zero bit offsets in both operands.

The benchmark registers 64 cases covering threshold boundaries and representative small, large, sparse, and dense inputs.

Are these changes tested?

Yes. The benchmark compiles on main, and Criterion lists all 64 new cases.

cargo bench -p parquet --bench row_selector --no-run
cargo bench -p parquet --bench row_selector -- mask_and_then_ --list

Are there any user-facing changes?

No.