A Simple Proof of the Universality of Invariant/Equivariant Graph Neural Networks