openai/gpt-chat-latest → B (5:1)

Side B fixes a real concurrency bug by shortening the lifetime of `tokio::sync::RwLock` read guards before later `read().await`/`write().await` calls in `handle_rpc_batch`, preventing self-deadlock during room creation and grants. It also adds an integration test covering room creation and improves test infrastructure with timeouts and optional log redirection, whereas Side A mainly adds a migration plan document and a thin `RouteContext` wrapper around `ThreadNav` without changing application behavior.