Boundary-Aware Uncertainty for Feature Attribution Explainers