Language Model Council: Benchmarking Foundation Models on Highly Subjective Tasks by Consensus | Synapse