[backend] drop useless check for existing workers
    
    This seems to be buggy.  Frontend never returns job which is
    already assigned to some worker, unless there's some bug..
    
    Removing this check allows us to actually properly react on SSH
    connection failure (either sshd died, or whole VM is unusable); so
    that Worker (a) terminates the VM and (b) asks Frontend to
    resubmit the job again.