Consolidation via Policy Information Regularization in Deep RL for Multi-Agent Games