Learned a new #Linux thing:

convmv
(Corrected)

and as a side effect, that there are different "normalization forms" in unicode.