The blog (and layout) is very neat. Btw you could simply pass nu_z to var_names for it to be resampled in the posterior predictive function. No need to extend the model with a new variable or move the sigma into it.
Your example is a great use case of this functionality.