docs: update README — 4-tier routing, Groq LLM, integrated Prometheus, correct cost saving 96%

fix: add missing httpx import to inference.py (Grafana push NameError)

fix: add conversational_handler to frontend routing config and badge CSS

fix: add REGISTRY to prometheus_client imports (Grafana push NameError)

feat: add conversational pre-filter (tier 0) for greetings and farewells

fix: skip intent context hint for very low confidence LLM calls

fix: use llama-3.1-8b-instant as default Groq model

feat: add Groq provider to LLM client (free, no billing required)

fix: replace prometheus-remote-write (requires py3.11) with manual protobuf+snappy

feat: push metrics directly to Grafana Cloud from HF Space

fix: add prometheus-client to serving requirements

feat: add /metrics endpoint for production monitoring on HF Spaces

fix: correct Banking77 intent names mapping and add RAG for template_handler

fix: download model from HF Hub instead of COPY artifacts

initial deploy: SBERT+LinearSVC router with RAG + LLM