Shrinkage with Robustness: Log-Adjusted Priors for Sparse Signals