Improving Anomalous Sound Detection with Attribute-aware Representation from Domain-adaptive Pre-training