Provably Correct Automata Embeddings for Optimal Automata-Conditioned Reinforcement Learning