Ptycho: handle large datasets by swapping stacks between GPU and main memory
This would use the same approach as for holotomo: use at least 3 stacks of frames, one processed, one uploaded to the GPU, one offloaded. This will use two streams, one for computing and another for i/o