Implement 'Shape "loopsubdiv"'

[wahn]

The one and only test scene distributed with the C++ code contains a subdivision surface:

Shape "loopsubdiv"  "integer nlevels" [1 ] "integer indices" ...

Beside the problem that we can't handle Include statements yet (see issue #41), we need to implement CreateLoopSubdiv(...) on the Rust side and a bunch of helper functions/structs:

std::vector<std::shared_ptr<Shape>> CreateLoopSubdiv(const Transform *o2w,
                                                     const Transform *w2o,
                                                     bool reverseOrientation,
                                                     const ParamSet &params) {
    int nLevels = params.FindOneInt("levels",
                                    params.FindOneInt("nlevels", 3));
    int nps, nIndices;
    const int *vertexIndices = params.FindInt("indices", &nIndices);
    const Point3f *P = params.FindPoint3f("P", &nps);
    if (!vertexIndices) {
        Error("Vertex indices \"indices\" not provided for LoopSubdiv shape.");
        return std::vector<std::shared_ptr<Shape>>();
    }
    if (!P) {
        Error("Vertex positions \"P\" not provided for LoopSubdiv shape.");
        return std::vector<std::shared_ptr<Shape>>();
    }

    // don't actually use this for now...
    std::string scheme = params.FindOneString("scheme", "loop");
    return LoopSubdivide(o2w, w2o, reverseOrientation, nLevels, nIndices,
                         vertexIndices, nps, P);
}

> pwd
/.../github/pbrt-v3/src/shapes
> rg "struct SD"
loopsubdiv.cpp
43:struct SDFace;
44:struct SDVertex;
51:struct SDVertex {
64:struct SDFace {
96:struct SDEdge {

[wahn]

Here is a first result (with only one instance of the killeroo geometry) for Rust:

There is a difference to the C++ version though:

> imf_diff -f -d killeroo-simple-cpp.exr killeroo-simple-rust.exr diff.jpg
differing pixels:	  0.032% (155 of 490000)
average difference:	 18.131%
maximum difference:	 67.999%
Summary: A few pixels differ strongly.
== "killeroo-simple-cpp.exr" and "killeroo-simple-rust.exr" are different

[wahn]

After commit a7c7382 (and using the bdpt integrator with 256 pixelsamples) we get almost identical results:

> imf_diff -f -d killeroo-simple-cpp.exr killeroo-simple-rust.exr diff.jpg
differing pixels:	  0.003% (14 of 490000)
average difference:	  1.581%
maximum difference:	  2.609%
Summary: A few pixels differ slightly.
== "killeroo-simple-cpp.exr" and "killeroo-simple-rust.exr" are similar