openai/gpt-chat-latest → B (4:1)

Side B fixes a substantive concurrency bug by ensuring Tokio RwLock read guards are dropped before nested read/write awaits in RPC handlers, explicitly preventing deadlocks during room creation and grant operations. It also adds an integration test for room creation and improves test reliability with timeout/logging changes, whereas Side A mainly exposes existing connectivity statistics in CLI output with formatting and unit tests but does not change core behavior.