Non-linear random effects models with continuous time autoregressive errors: a Bayesian approach