fix: use half-up rounding in raw value calculation

This approach reduces the frequency of rounding errors where repeated addition and subtraction operations are supposed to cancel each other out, but don't, because the absolute brightness value is always rounded down. I.e. when you run brillo -A 5 and brillo -U 5 you should always get the initial value, and this PR brings the tool closer to that goal. Partially fixes #5 (closed)