Given a vector of multivariate normal data, a matrix of covariates and the data covariance matrix, generate new multivariate normal samples that have the same covariance matrix based on permutations of the transformed data residuals.