Improve foliage shadow culling per split CPU performance by avoiding DrawMeshInstancedIndirect, just submit RenderMeshIndirect and cull per shadow caster up front