Implement multi-participant real-time chat backend

[AlbertCorry]

Summary

NLWeb currently supports single-user conversational search via /ask, but there is no backend for real-time multi-participant chat rooms where multiple humans and an AI assistant exchange messages concurrently.

Expected behavior

• Users can create conversations, join/leave rooms, and send messages over WebSocket
• AI responses are generated through the existing NLWeb query pipeline (handler wrapped, not modified)
• Messages receive strictly ordered sequence IDs under concurrent writes
• Single human + AI: direct routing to AI (no broadcast overhead)
• Multiple humans: broadcast to all participants; AI receives context from recent human and AI messages
• REST endpoints for conversation lifecycle; WebSocket for real-time messaging
• Per-conversation queue limits and participant capacity with appropriate HTTP errors (e.g. 429)
• In-memory storage backend for development/testing

Context

The repository contains extensive test specifications and API documentation for this chat system under tests/ and .claude/, but the AskAgent/python/chat/ implementation is not yet present.