Extreme compression of sentence-transformer ranker models: faster inference, longer battery life, and less storage on edge devices.