Challenges in Markov Chain Monte Carlo for Bayesian Neural Networks