Unifying GANs and Score-Based Diffusion as Generative Particle Models