FIx CUDA performance bug
The updated CUDA host memory management did not pass the host pointers for the visibilities and uvw data to the reuse_memory function. This resulted in slow memory transfers and non-overlapped io/compute in degridding.
The updated CUDA host memory management did not pass the host pointers for the visibilities and uvw data to the reuse_memory function. This resulted in slow memory transfers and non-overlapped io/compute in degridding.