- read header from .nii and not just .nii.gz
- detect duplicates
- organize installing dependencies so that things like MNE and nibabel do not need to be installed as they are only used for updating the listing
- make functions private in cohort_creator/data/_update.py