separate self-collision meshes from environment-collision meshes (BIG)

[rdiankov]

current method doesn't allow robot to be padded because it will start affecting self-collision results.

The solution is to have two meshes:
self-collision mesh is high-res, environment collision mesh gets convex decomposition.

collision checkers need to put self-collision stuff into a separate space away from the "real-world"

find a way to optimize collision checking for convex decompositions