Unsupervised learning of acquisition variability in structural connectomes via hybrid latent space modeling