Set job child process default scheduler to PB

at fork (and not later on as before) which fixes a bug where child
processes that have been initially run by the CFS would somehow remain
in its queue and would get scheduled even though they are dead which
causes critical errors.