Split Algorithm_SORT kernels into comparison and non-comparison #551
Description
Currently we use a comparison based sort for Seq and Omp, but radix sort for hip and cuda. Given the difference in complexity with fixed size data these should be different kernels so the attributes can be accurate.