The Cache, Route, Compress Playbook: Cutting LLM Costs 70–90% in Production