Sorry that you had to add it @alteredq.

I'll update all the examples in a sec.

Refactored wireframe example to use single material for both triangles and quads, added geometry with mixed quads and triangles.

Added wireframe materials example (using lines, triangles and quads). Thanks to @jwagner and @pyalot

This allows to give unique values for each face use of the vertex (as opposed to unique custom attributes per vertex or per face).

For the moment just Vector3 and Color attribute types are handled.

Primary motivation for this binding type is to be able to do barycentric wireframe rendering in shader.

Because sometimes you just need to get rid of junk stuff sneaking into asset pipeline.

Don't merge yet, still needs testing.

Displacement should be computed in object space (and shouldn't need scaling).

Correct shadow casting still needs a custom depth material but at least shadow receiving should now work with displaced vertices.

This missing parameter was most likely the cause of anisotropy troubles on Linux - it caused "undefined" to be sent to GPU and then later on with Math.min(aniso, maxAniso) it started to send "NaN" instead.

Don't know anymore what's going on - preventing setting anisotropy to 1 in all cases apparently makes failures on Linux go away but this still doesn't explain why original anisotropy example works ok with setting anisotropy to 1.

Please note that there is now extra logic to allow for resetting anisotropy back to 1 (for platforms that don't exhibit this troubles), if we just prevented setting anisotropy to 1 in all cases, reset wouldn't be possible.

Phew! Thanks @mrdoob for valiant testing.

This makes it a bit more heavy but saves quite a few varyings, making it possible to handle more lights in ANGLE.

Fixes #2260