intel/fs: work around gen12 lower-precision source modifier limitation