openai/gpt-chat-latest → A (3:2)

Side A makes a broad, lasting type-safety refactor by changing `resolve_item` to return `CanonicalItemUrl`, propagating canonical URL newtypes through validation, ranking, RPCs, and connectivity logic, and adding `Deref<Target=str>` implementations to reduce string conversions. Side B fixes important test infrastructure for OAuth E2E flows (correct request-body reading, safer query/token parsing, redirect handling, and mock error handling), but those improvements are confined to test mocks rather than the project's core APIs and data model.