* Use a slightly better (but still not exact) approximation for the view
vector when pre-shading the light cache. This still doesn't give exactly the
same results as non-light-cache shading, but it's better. Will investigate
getting a better view vector when there's more time - or if anyone has a
simple formula to derive shi->view from shi->co that would be great to
hear about too :)
8d5c14b20d