Additionally, this change introduces a M_CHARCONV state flag which
gives us some more control about when character set conversions are
actually done.  Current versions of mutt would happily apply
character set conversions when, e.g., saving a text/plain attachment
to a file.  (We had at least one corrupt russing translation file
due to this mis-feature.)

Additionally, we clean up some of the character set related code in
handler.c.  Most of that is now done by the decoder functions in
charset.c.