load_checkpoint.sh logins for replicas fail if p4d starts slow with systemd.
The systemd start mechanism is a 'fire and forget' style. That is, there can
be a delay between the time the 'systemctl start p4d_N' call returns, and when
the system is actually up (e.g. responsive to 'p4 info').
When used to seed replicas, load_checkpoint.sh fires off a 'p4login -v -service'
call, which fails with TCP connect errors if the p4d service doesn't start
A possible fix could be to use the start_p4d() function, which returns
only when p4d starts.