S2DM: Sector-Shaped Diffusion Models for Video Generation