openai/gpt-chat-latest → A (4:1)

Side A fixes multiple concrete failures in the OAuth test infrastructure: it corrects query parsing (`str/split` with regex), reads POST bodies from `getRequestBody`, guards against null tokens/state, adjusts redirect handling, wraps handlers to avoid crashes, and improves Playwright test synchronization and selector lookup. Side B is largely a storage refactor from deque to list with schema changes and query-time capping, but it mostly changes implementation strategy rather than addressing a demonstrated correctness issue, making its lasting impact less certain than A's direct restoration of broken end-to-end authentication tests.