There are two major changes here:
1) Previously I was inverting the winding of verticies only when calculating the normal. This was the wrong approach and I instead needed to invert the order of the verticies before calculating the normal. This is because most STL processing software (eg: Slic3r) will recalculate the normals based off the winding (correctly).
2) I *think* I am doing proper BFC handling now, although TBH the BFC handling logic in ldraw is super complex to me. See: https://www.ldraw.org/article/415 ("Rendering Engine Guidelines").
There was bad behaviour if there were nested include files combined with a recursive invert flag, resulting in incorrect surface normals for sub-parts. For example try ldraw/parts/3020.dat before this commit. This is not the best way to solve the problem, but it may suffice for now. IMO the best way would be to "flip" the invert flag between states, rather than explcitely set it to 1 or 0. That way recursively setting invert would simply flip the flag.
