Improving the Diffusability of Autoencoders