VTK-m can now defer de-allocations of CUDA memory.
This is done as cudaFree blocks execution on all cuda streams. When doing operations such as moving virtuals to CUDA memory space we want to defer the cuda free as long as possible to not block other stream execution.