Replaced matrix related FFP code with explicit matrix calculations. The modelview matrices now get uploaded at a single point via glLoadMatrix4 rather than using the GL1 matrix stack. In the GL2 variant this will become a uniform upload.