We integrate acceleration to get velocity, but that means sitting on the pad
for a long time can add substantial error to the velocity value. Each
second, take the velocity value from a full second ago and subtract that out
of the current velocity. Once we detect boost, this will stop, which means
that as long as we detect boost within a second, we won't have subtracted
out any "real" velocity.
This keeps the pad velocity hovering around zero, which is pretty useful.
Signed-off-by: Keith Packard <keithp@keithp.com>