Using Time-Aware Graph Neural Networks to Predict Temporal Centralities in Dynamic Graphs