Should we ensure transform is rigid on image load?

[jdtournier]

As discussed on #1363.

Suggestion to add checks to verify that the transform is rigid-body, and actually modify it to the nearest rigid transform on load (commit 0f41e46), reverted in 2f9c558). It's not obvious what to do here, especially if we want to handle NIfTI consistently (qform can only represent a rigid transform). Would this break the registration apps?

[maxpietsch]

Currently, we warn when transforming the header with a non-orthonormal matrix and when viewing the image, if the voxel spacing is inconsistent.

mrtransform a.mif -linear Affines.txt at.nii  # [WARNING] the input linear transform is not orthonormal
mrview b.mif -overlay.load at.nii -overlay.opacity 0.3 # [WARNING] voxel spacings inconsistent

The above overlay uses the wrong transformation but It works for .mif files.

For Nifits, it might make sense to force rigid transformations in the header but I would not get rid of this feature for .mif. But this means we'd need to regrid .mif images when converting them to .nii. Also, I think we are not the only software package that allows applying non-rigid transformations to the header to avoid resampling. So, I'd keep things as they are, given that the user has been warned...