Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion