Integrating Random Effects in Variational Autoencoders for Dimensionality Reduction of Correlated Data