Allow NIfTI cohort files without source_mask_file if group mask is provided

[tsalo]

This will require documentation and a warning, since the group mask and individual mask files are typically complementary. If users only use a group mask without accounting for run-wise differences in coverage, they may end up with NaNs or zeros throughout their ModelArray arrays, so this should be done carefully.

[araikes]

On the NIFTI side of things, both masks types have value. Subject-specific masks define NaN elements that are then used to for thresholding and deriving the analytic mask (which, incidentally, should probably be returned by ModelArray). This should probably be preferred the preferred behavior. Having the "group mask file" intuitively suggests that it's the analytic mask.

Maybe a better approach is to still require the "source_mask_file"and document that the subject mask files ultimately define the analytic mask (depending on threshold value in ModelArray) while the group mask file defines the header information for writing files.

You can still have the case example of having all subject masks be the group mask file (which isn't unreasonable if you've derived a study specific template and that has an already thresholded mask, which is my mouse brain use case more often than not), but with an example that highlights that this may not be the only use case.