Add HIP Launch policy with blocks per thread option. #1305
Description
Need something for HIP that is analogous to cuda_launch_t<async, numthreads>. This was not implemented for HIP due to no solution for blocks_per_sm, but there is a use case for this in transport code.