two reasons for this. One, it saves a bunch of boilerplate. Two, it squashes unicode to IDNA-in-a-`str` (even on python 3) in a way that it turns out we rely on to give consistent behaviour between python 2 and 3.
