Optimize `compare_greater` for faster parquet writer

@jhorstmann notes:

>
> I looked at this function in #7916 and noted
>
> I looked into also specializing the `get_min_max` function by moving the logical type checks outside of the loop, but that did not bring any further measurable improvement.
>
> Maybe trying again to vectorize that code could be worthwhile.

_Originally posted by @jhorstmann in [#8783](https://github.com/apache/arrow-rs/issues/8783#issuecomment-3490077327)_

Locate `compare_greater` in the parquet writer and inspect the `get_min_max` loop and its logical type checks. Compare any vectorization attempt with the current implementation using the project's existing performance measurements; done means a measurable writer speedup without changing output.

Written by the indexing model from the issue text.