Fix mutt_parse_crypt_hdr() sign_as behavior.

Fix mutt_parse_crypt_hdr() sign_as behavior.

Currently, if a message is postponed with only the E encryption flag,
when it is resumed, pgp_sign_as will be overwritten (with an empty
string).

This intention of the behavior was probably to restore the exact state
of pgp_sign_as upon resuming.  However, if the message wasn't marked for
signing, the state of pgp_sign_as is not known.  This patch changes the
mutt_parse_crypt_hdr() to only set an empty pgp_sign_as if there is an S
flag (and if set_empty_signas is true).

mutt_edit_headers() also uses the function, but it doesn't want to
overwrite pgp_sign_as with just an S flag.  The set_signas parameter is
renamed to (a hopefully clearer) "set_empty_signas".