Mask-Weighted Spatial Likelihood Coding for Speaker-Independent Joint Localization and Mask Estimation