Allow for multi-module detectors to process data separately per module

[tguruswamy]

For some multi-module Lambda systems such as FLEX, it makes more sense to allow the user to process each module's data separately, rather than as one stitched image. This could be made possible by stacking module images along an extra NDArray dimension. This would make it easier to use ROIs to select a particular module or slice the data for only a particular module when saving into HDF5, for example.