Add benchmarks for bounding WKB arrays

[paleolimbot]

This is not an exhaustive search for the best option, just a check of the Parquet implementation to see where we're at.

Three situations with an identical coordinate count are considered:

• A raw (aligned) double array. This should be the fastest/simplest (although isn't somehow).
• A single linestring (where parsing overhead should be minimal)
• All coordinates as their own WKB points. Here, parsing overhead dominates

I think the conclusion here is that the current Parquet implementation should be simplified to something more like what we have in geoarrow-c.

(The non-Parquet versions are probably optimistic because they don't check Z and M)

CPU Caches:
  L1 Data 64 KiB
  L1 Instruction 128 KiB
  L2 Unified 4096 KiB (x12)
Load Average: 1.84, 1.99, 1.88
-----------------------------------------------------------------------------------------------------
Benchmark                                           Time             CPU   Iterations UserCounters...
-----------------------------------------------------------------------------------------------------
BenchBoundDoubles                              911401 ns       910030 ns          764 items_per_second=1.09886G/s
BenchBoundDoublesWithEmptyCheck                702162 ns       700760 ns         1029 items_per_second=1.42702G/s
BenchBoundWKBLinestringUsingGeoArrowHpp        599640 ns       598843 ns         1160 items_per_second=1.66989G/s
BenchBoundWKBLinestringUsingParquetBounder    1295315 ns      1280583 ns          542 items_per_second=780.894M/s
BenchBoundWKBPointsUsingGeoArrowHpp           6901854 ns      6892653 ns          101 items_per_second=145.082M/s
BenchBoundWKBPointsUsingParquetBounder        8595789 ns      8583280 ns           82 items_per_second=116.506M/s
BenchBoundWKBPointsUsingUnalignedSequence     1270639 ns      1268723 ns          541 items_per_second=788.194M/s