Parametric Shadow Control for Portrait Generation in Text-to-Image Diffusion Models