On the Robustness of Post-hoc GNN Explainers to Label Noise