DUAL: Textless Spoken Question Answering with Speech Discrete Unit Adaptive Learning