AFD-SLU: Adaptive Feature Distillation for Spoken Language Understanding