Training Neural Samplers with Reverse Diffusive KL Divergence