A comprehensive study of spike and slab shrinkage priors for structurally sparse Bayesian neural networks