Use `contiguous_split` for shuffling

[seberg]

legate-dataframe uses the same pattern as rapidsmpf was using, but this is now changed here: rapidsai/rapidsmpf#371

We should do the same refactor here, should make the code slightly simpler and faster.