Optimize recommendation latency with caching

Implement Redis caching layer to reduce recommendation generation time. Target <50ms p95 latency.