Multimodal Prediction of Pathological Complete Response in Esophageal Cancer Using Automated Machine Learning and Variational Autoencoder-Based Synthetic Data Augmentation