Scalable Bayesian Inference for Excitatory Point Process Networks