tnl: Push down the gl_vertex_array inputs into tnl drivers.