removed the unused `cuda_block_size` for the `gpu.generate_benchmark`

function. For now the only way to set the cuda_block_size size is to use
pass it in the `ps.KernelConfig` to the generated kernel