Obtaining error-minimizing estimates and universal entry-wise error bounds for low-rank matrix completion