Unsupervised Data Selection via Discrete Speech Representation for ASR