LiteVAE: Lightweight and Efficient Variational Autoencoders for Latent Diffusion Models