Data Augmentation for Supervised Graph Outlier Detection with Latent Diffusion Models