[C++] Improve kernel performance on small batches

It seems that invoking some kernels on smallish batches has quite an overhead:
```Java

ArrayArrayKernel/32768/100 2860 ns 2859 ns 245195 bytes_per_second=10.6727G/s items_per_second=2.86494G/s null_percent=1 size=32.768k
ArrayArrayKernel/32768/0 2752 ns 2751 ns 249316 bytes_per_second=11.093G/s items_per_second=2.97775G/s null_percent=0 size=32.768k
ArrayArrayKernel/524288/100 18633 ns 18630 ns 36548 bytes_per_second=26.2097G/s items_per_second=7.03561G/s null_percent=1 size=524.288k
ArrayArrayKernel/524288/0 18260 ns 18257 ns 38245 bytes_per_second=26.7451G/s items_per_second=7.17933G/s null_percent=0 size=524.288k
```

We should investigate and try to lighten the overhead.

**Reporter**: [Antoine Pitrou](https://issues.apache.org/jira/browse/ARROW-10026) / @pitrou
#### Related issues:
- [[C++] Improve array expression and kernel evaluation performance on small inputs](https://github.com/apache/arrow/issues/20277) (relates to)

_{**Note**: *This issue was originally created as [ARROW-10026](https://issues.apache.org/jira/browse/ARROW-10026). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}

Start by locating the ArrayArrayKernel benchmark represented in the issue and compare the small-batch results with the larger-batch cases. Investigate where kernel invocation overhead is introduced; done means demonstrating lighter overhead on small inputs with updated benchmark results, though the issue gives no target or test file.

Written by the indexing model from the issue text.