This patch caches VAO objects instead of re-emiting all pointers per draw call.
Configuring this pointers is known as a fast task, but it yields too many GL
calls. So for better performance, just bind the VAO instead of 16 pointers.