Comparing Classical and Deep Learning Approaches for Early Sepsis Prediction with Model-Specific Explainability