openai/gpt-chat-latest → A (3:2)

Side A restores broken end-to-end authentication tests by fixing multiple concrete defects in the mock OAuth servers: correct query parsing with regex splitting, reading POST bodies from `getRequestBody`, guarding null bearer tokens/state, using proper redirect response handling, and wrapping handlers to avoid crashes, alongside improving Playwright test selectors and synchronization. Side B adds a useful UI enhancement by making ranked-list pin icons submit the existing `set_garden_pin` unpin flow with supporting CSS and browser tests, but it is a narrower feature addition than A's broad reliability fixes for the project's authentication test infrastructure.