ReGVD: Revisiting Graph Neural Networks for Vulnerability Detection