Data Similarity-Based One-Shot Clustering for Multi-Task Hierarchical Federated Learning