I wonder if these concerns could be addressed by defining the sort key
as equal to the addr-name sent by the server in response to
FETCH-ENVELOPE. IIRC that's still 2047-encoded, but "like the
addr-name you send for FETCH ENVELOPE, but 2047-decoded" is realistic
and still simple.
This is completely reasonable, and also very much in line with the
rest of the base SORT draft. Give me a bit of time and I'll try to
rework the draft's wording to phrase it in terms that line up with the
FETCH ENVELOPE response.