Thanks for reporting this. It’s probably good that newer versions of ITK are more careful about interpreting nifti variants, as this has been an ongoing source of confusion. I don’t think we should push back in that, but instead offer a more robust nifti loader that as you suggest is more user-friendly in pointing out bad data and offering loading solutions. This would be a chance to handle other variants nifti types, like time series.
I did a prototype of a python-based reader for diffusion MRI that could be used as a starting point. It would be great if people who rely on nifti in their workflows could pick up and generalize this to support their needs. Perhaps even a dedicated module like RawImageGuess or ImageStacks would help people deal with some not-uncommon issues like left-right flip, qform/sform disagreement, or making use of intent types.