Unconditional Human Motion and Shape Generation via Balanced Score-Based Diffusion