Turns out the code for div by a power of 2 is already almost capable of optimizing this to a shift-by-0 or and-with-0, which will be further folded into nothingness; let's beef it up to handle div by 1 as well. Fixes #2820.