openai/gpt-chat-latest → A (3:2)

Side A replaces synchronous vote handling with a dedicated settlement worker that batches event logging and ranking recomputation, introduces cached ranking reads (`ranked_items_cached`) to avoid recomputing on every HTTP render, and removes obsolete demo-counter code and locking. Side B fixes canonical ontology root/path normalization and related garden rendering bugs with strong regression tests, but its impact is narrower than A's broader performance and architecture improvements.