Preemptively Pruning Clever-Hans Strategies in Deep Neural Networks