Resource Heterogeneity-Aware and Utilization-Enhanced Scheduling for Deep Learning Clusters