Add observability, Gemini support, and frontend chat UX improvements

[azmuth13]

Summary

This PR improves the AI Observability Agent across backend reliability, model support, diagnostics, testing, and frontend UX.

It adds structured observability to the backend, introduces configurable Gemini support alongside Groq, improves graceful error handling for LLM/provider failures, and upgrades the React chat UI with better response rendering, multi-session chat history, and a lighter polished design.

What Changed

Backend

• added structured JSON logging and request IDs
• added /api/debug/status diagnostics endpoint
• exposed active LLM provider/model in debug and chat responses
• added configurable LLM provider selection (groq or google)
• added Gemini support using langchain-google-genai
• improved graceful handling for connectivity/auth/quota failures from LLM providers
• added runtime flags such as mock mode, Pinecone toggle, and MCP toggle
• added MCP config and improved MCP docs

Agent / Retrieval

• added stage latency tracking across graph steps
• added tool usage, retrieval source, retrieval hit, and evidence metadata
• normalized Gemini response content so provider-specific content blocks serialize correctly
• added retrieval source reporting and local fallback awareness

Frontend

• improved assistant answer rendering into readable sections
• added metadata display for intent, tool usage, retrieval, latency, and active model
• added persistent chat history with localStorage
• added multiple chat sessions and ability to start a fresh chat context
• added clear chat action
• added message timestamps
• added follow-up suggestion chips
• redesigned the UI with a lighter, more polished visual style

Testing / Evaluation

• added route tests for /api/chat
• added diagnostics endpoint tests
• added log tool tests
• added evaluation dataset and validation test
• updated smoke/health test setup

Why

These changes make the project more production-shaped and easier to demo, debug, and extend:

• better observability for an observability-focused agent
• easier switching between Groq and Gemini
• clearer UX for end users
• better resilience when external model providers fail
• stronger regression safety through tests and evaluation inputs