Semi-Supervised Nonlinear Distance Metric Learning via Forests of Max-Margin Cluster Hierarchies