HiKG-DDI: Pretrained model checkpoints (v1.0)

Pre-trained PyTorch checkpoints for HiKG-DDI at every main configuration reported in the accompanying paper: DDI-Bench S0 (5 seeds, best single 95.28 F1), DDI-Bench S1 cold-start (5 seeds), DDI-Bench S2 cold-start (5 seeds), and TWOSIDES S0 (5 seeds). The top-3 ensemble on DDI-Bench S0 achieves 95.49 F1 (+1.09 over EmerGNN, the prior SOTA). Each checkpoint includes the model state_dict, training hyperparameters, and validation metrics. Requires the companion features bundle for inference. Associated code: https://github.com/Omarmahmoud711/Khemical