In the pam_krb5 config, instead of just skipping the pam_echo in case
of failure, immediately die.
This works around a bug in pam-auth-update where default=1 is treated
differently depending on whether or not the "end" in success=end has
been replaced with a number. This was causing pam-auth-update to
spuriously detect changes to /etc/pam.d/common-auth.
This change does have the effect that a failure of pam_krb5 will no
longer bubble down to any other potential auth providers. However, I
think that the scenario of (a) using >=debathena-login, (b) having a
second PAM auth module you want to use that, (c) is managed by
pam-auth-update and not by hand and (d) comes after pam_krb5 in
pam-auth-update's ordering scheme is pretty unlikely.