Note that for this to work with nnatom in the current upstream you'll
also need (ahem!) my patch in bug#65467: bug#71888 must be responsible.

> Without diving into this right now, it seems like this is something that
> should be governed by the nnmail abstract backend, from which nnml and
> friends inherit. I would dearly, dearly love it if all backends that
> might potentially create an on-disk directory from a group name would
> use the same code (applying the same user options) to do it, essentially
> transparently. It makes me nervous when various functions in various
> places repeat similar-but-not-quite-identical routines in encoding and
> decoding group names. I suppose that URL encoding/decoding functions
> might end up being an okay tool, but I wonder if Elisp doesn't already
> have some prior art here. I'll do a bit of reading.

That's worthwhile of course, but here, for the time being, I've decided
I'm only grappling with the new allowance of '/'s in group names. :-)

(A further improvement involves replacing the '/'s in the code with
'-directory-separator-character', but that's for another report)

Regards,
James