Learning Path Distributions Using Nonequilibrium Diffusion Networks