openai/gpt-chat-latest → A (3:1)

Side A fixes concrete failures in the OAuth test infrastructure by correcting query parsing (`str/split` with regex), reading POST bodies from `getRequestBody`, guarding null bearer tokens and state values, wrapping mock handlers in error handling, and updating the Playwright helpers to use real selectors instead of brittle timing. These changes directly restore broken end-to-end authentication flows, whereas Side B mainly adds a CLI path for room creation and simplifies room metadata by removing unused visibility handling, which is useful but less critical and less of a correctness fix.