Text this: Benchmarking TabNet, NODE, and FT-Transformer for Software Defect Prediction: An Empirical Comparison and Explainability Analysis