Overlap-aware End-to-End Supervised Hierarchical Graph Clustering for Speaker Diarization