Recent advances in domain decomposition methods for large-scale saddle point problems