If I create a new nnml group with a non-ASCII name:

   `G m Æblegrød RET nnml RET'

The group turns up in the Group buffer like this

   �blegr�d

The problem is of course that Gnus thinks the name is unibyte (utf-8
specifically).

We could fix it by setting `gnus-group-name-charset-group-alist' to
nil, but then names of nntp groups which really _are_ unibyte would be
wrong.

I think it's a big mess that the list of newsgroups
(`gnus-newsrc-alist', `gnus-newsrc-hashtb' etc.) can contain a mix of
unibyte and multibyte names.  It also breaks `j'
(gnus-group-jump-to-group) in the Group buffer and probably more
stuff.

I don't see any easy solution, though.