machined: keep track of the initial leader PID of a machine

This way we can without races always determine the machine for a leader
PID. This allows machine managers to query the machine for a forked off
container/VM  without a race where the child might already have died
before we could read the cgroup information from /proc/$PID/cgroup.