We ended up meeting today. There seems to be consensus on the following:
Ideally, the existing f=f can continue, with additional
clarifications: if the character on each side of the SP CRLF sequence
is 7-bit US ASCII, go ahead and delete the CRLF but leave the space
(as is done now). If either character is not 7-bit US ASCII, then
use language-specific folding rules to decide if the SP should be
deleted or not. Since the point of f=f is to permit line wrapping,
the client needs language-specific rules anyway. It just needs to
now apply them during f=f decoding as well.
If there is violent objection to the above, the alternate solution is
a new format parameter (content-type=text/plain; format=wrap) with
the rule that a SP CRLF is inserted as desired, and deleted on
receipt. The current f=f would be depricated.