Default CPU workgroupsize can be inadequat for higher-dimensionsal kernels

Created by: vchuravy

As observed in https://github.com/WaterLily-jl/WaterLily.jl/pull/133#issuecomment-2263500896

When we have a grid like (64, 64, 64) using a WG size of (1024, 1, 1) is inefficient. Instead WL uses (64,1,1) and maybe KA should "spread" the 1024 to (64, 64,1).