float to int now takes place in a place which is offset by m_normalizedFront,
so that proximity to zero is avoided, which would have failed to drop
accumulated errors as effectively. I still have some work to do, esp. in the
code that interpolates colors, as the color components need the same treatment.
Polygon filling also needs to be updated eventually. Right now both worlds can
coexist because I have parallel sets of variables or functions.
