Formal derivation of Mesh Neural Networks with their Forward-Only gradient Propagation