LLM Result Cache

Tiny dependency-free TTL cache that skips a repeat LLM/API call entirely when the same input recurs. Fixes a common cost leak in agents that re-score/re-analyze the same URL, document, or input across separate tasks.