pub fn downshift(image: &mut [i32], d: u32)
Downshift: multiply each sample by 2^-d, rounding to nearest integer (Section 3.2.3, Eq. 13).