Approximate blocked Gibbs sampling for Bayesian neural networks