Make source_file flexible

[araikes]

Instead of hardcoding the source_file column, allow users to specify the column. This will make it easy to extend ModelArray to allow multiple .h5 inputs for multi-scalar analyses.

[tsalo]

I think there is support for wide-form cohort.csv files where you can use --scalar-columns, but it only seems to be implemented for modelarrayio cifti-to-h5.

EDIT: Sorry, I misread your comment. The --scalar-columns argument offers different functionality to what you're suggesting.

[tsalo]

I think --scalar-columns actually does address the problem. Is this good to close now that #34 is merged?