Add per-request LLM provider toggle (OpenAI / Gemini) + React frontend

SHL recommender — initial deploy