Towards Prototype-Based Self-Explainable Graph Neural Network