Add retry with exponential backoff for rate limits; add delay between samples

Fix target to call LLM directly (no sandbox dependency); reduce to 3 golden samples

Initial commit: LLM-as-a-Judge evaluator