Interpretable Deep Learning Paradigm for Airborne Transient Electromagnetic Inversion