Computationally Efficient Distribution Theory for Bayesian Inference of High-Dimensional Dependent Count-Valued Data