Investigate inconsistent raster subsetting performance

[jpswinski]

When running the rasterized subsetting benchmark on v4.0.2 (which does not have the optimized GeoJsonRaster code) we consistently see timing numbers between 16 and 19 seconds.

When running on @elidwa's development machine, the benchmark on main 5fc9711 (which also does not have the optimized GeoJsonRaster code) runs in just under 8 seconds, while the benchmark using the optimized GeoJsonCode that is optimized d1e776b runs in just over 6 seconds.

When running on @jpswinski's development machine (which is an identical instance type to @elidwa's), we see the optimized code run in just under 4 seconds.

We need to investigate why there is this large disparity between the benchmarking runs.