arrow: add oversized coalesce take benchmarks (#9799)

## Summary
- add `coalesce_kernels` take benchmarks alongside the existing filter
coverage for primitive, view, utf8, and dictionary schemas
- add oversized repeated-index benchmark cases that stress the
materialized `push_batch_with_indices` fallback with a configured
`biggest_coalesce_batch_size`
- extend the benchmark harness so take benchmarks can use explicit
output lengths and optionally drain all completed batches while running

## Verification
- cargo fmt -p arrow
- cargo clippy -p arrow --bench coalesce_kernels --features test_utils
-- -D warnings
- cargo bench -p arrow --bench coalesce_kernels --features test_utils --
'extra_large_repeat'

Signed-off-by: 蔡略 <cailue@apache.org>

Author: ClSlaid