Reward-Guided Iterative Refinement in Diffusion Models at Test-Time with Applications to Protein and DNA Design