perf: Optimize Latency (Remove query translation, async cache, tune LLM threads, limit tokens)

feat: Implement Phase 1 - Semantic Cache, Observability Logging, and Citation Verification