The blur of 1.0 is invariant, so that it should be preserved when scaling. Using `blur * scale` also scales the invariant, which is wrong. Tests show that `(blur - 1) * scale + 1` matches FP's behavior.