The division functions included in this crate pop the link-register value directly into pc to return. This is correct for interworking code with v5T and later, but it is incorrect for v4T.
With v4T, you must use bx to switch code modes.
At this time I'm not aware of a way to cfg on the ARM version of a target. I think the best fix is to just declare the functions as weak linkage, and then v4T users can just override the functions on their end.