Bayesian robust designs for linear models with possible bias and correlated errors