openai/gpt-chat-latest → A (4:1)

Side A repairs the OAuth test infrastructure with concrete correctness fixes: it corrects query parsing (`str/split` with regex), reads POST bodies from `getRequestBody`, avoids null handling crashes, fixes redirect/state handling, wraps the mock handler to prevent server crashes, and updates Playwright helpers to use real selectors and deterministic waits. Side B contains a mix of UX and routing changes (renaming `/vote/compare` to `/vote`, removing the swap button, changing fallback behavior, and passing a delegate option), but those are largely feature and cleanup changes rather than restoring broken core test functionality.