Approximately 5-10× faster guide deformation.

Over 5× faster long-hair generation and over 2× faster short-hair generation.

Over 2× faster hair clump computation.

Production-inspired guide deform methods that either use existing hairgen weights or compute barycentric weights for very stable and smooth interpolation.

Guide primitive attributes can now be types other than float. For non-float attributes, a hair will get the value of the closest guide.