Improving Sustainability of Adversarial Examples in Class-Incremental Learning