New aperture model: adaptive column

[lctseng]

Currently all columns (LFs) are stored in GPU memory.
Need to determine some way that can adaptive select the columns that should be in GPU memory

Maybe we should find out the projection that cover the data cameras,
so that we can determine which columns to transfer