Remove empty cells from MEX files before loading them

[arteymix]

This is causing problems downstream because we end up storing a lot of text for the cell IDs and indices for cell type assignments.

[arteymix]

No package out there that read 10X MEX supports writing it back to disk, so we will have to include that logic in the loader.

The simplest solution I could come up with was to read all the used column indices and rewrite the sparse matrix by shifting the column indices. The reported cell IDs also have to reflect that, so getSingleCellDimension() will have to peek at the matrix file.

[arteymix]

Completed! I'll deploy it tomorrow.