Multi-modal Variational Autoencoders for normative modelling across multiple imaging modalities