Diffusion-Based Speech Enhancement in Matched and Mismatched Conditions Using a Heun-Based Sampler