Variational Approach for Efficient KL Divergence Estimation in Dirichlet Mixture Models