Fully Bayesian inference for neural models with negative-binomial spiking