Approximation of Laws of Random Probabilities by Mixtures of Dirichlet Distributions with Applications to Nonparametric Bayesian Inference