D3PO: Preference-Based Alignment of Discrete Diffusion Models