Fortytwo: Swarm Inference with Peer-Ranked Consensus