'Random', without qualification, is commonly taken to mean 'with uniform distribution'. Otherwise it has no specific meaning and could well be a synonym for 'arbitrary' or 'haphazard'.

The behavior reported is buggy and in my opinion should be fixed if possible. I have done simulation research in the past and do not consider them minor. If I had results that depended on these functions, I might want to rerun with the fixed versions to make sure the end results were not affected. I would certainly want the fixed behavior for any future work.

I do not see any promise of reproducibility of sequences from version to version. I do not really see the point as one can rerun with the old Python version or copy the older random.py.

The old versions could be kept with with an 'old_' prefix and documented in a separate subsection that starts with "Do not use these buggy old versions of x and y in new code. They are only present for those who want to reproduce old sequences." But I wonder how many people would use them.