krb5 traces missing from journald logs at debug level 3

[simo5]

As discovered in #43 the way systemd starts daemons end up creating a redirection t a socket that breaks the ability to use '/dev/stderr' as a path name to be able to effectively reopen fd=2.

This means debug level 3 ends up not containing krb5 traces because the file open fails.
We should try to see if we can find a mechanism to deal with this, in the worst case creating a socket of our own and then redirecting its output back to stderr, or perhaps finding out if we can pass an FD number instead of a path to KRB5_TRACE, maybe asking krb5 to add a KRB5_TRACE_FD option ?