Hmm... I'll need to rewrite this one, the result seems to be wrong.

take 2: the 'fxaa' shader busts the transparency, so I'm setting the transparency to the transparency of the "middle" pixel instead. Changed after comment by alreredq - now the middle pixel is only fetched once.

wanted to fix the + 0.001 issue in `THREE.LatheGeometry` but ended up fixing, rewriting and cleaning up quite a bit of it. I think the UVs now looks better (using webgl_geometries.html) but UV behavior is slightly different so beware (although it should be difficult to support non-complete wrapped lathe UVs). UVs + mergeVertices + subD should be looked at another day, but that will be another issue.

making SubD code a little more readable. problems with UVs seems fixed. only the LatheGeometry subdivision example fails. other geometry seems to be working ok.

This is now done as method of Matrix3 object which gets as a parameter Matrix4 object to be inverted (to be consistent with how Matrix4 => Matrix4 inversion is done).

As per discussion in #1615.

Added some vars. The compiler is already optimising this so there is no point on doing this risky pattern anymore.

Matrix4's static methods makeFrustum, makePerspective, makeOrtho are now normal non-static public methods. Fixes #1615.