Describir: Conditional Variational Autoencoder for Learned Image Reconstruction