the one-second sleep between invocations was moved to after the actual
invocation. This sleep could be interrupted by a SIGCHLD in the case
of a failing service, and the service could be restarted way more than
the documented once-per-second. Move the sleep back before the fork(),
just don't sleep before the very first fork, so the service starts faster.
Silence by: Daniel J. Bernstein (no telling if he'll fix this in later
daemontools releases, so I guess we'll have to do it as part
of our port..)
Approved by: nbm (maintainer)
