openai/gpt-chat-latest → A (4:1)

Side A introduces a substantial architectural improvement by adding a settlement worker that batches vote persistence and ranking recomputation, separates cached ranking reads (`ranked_items_cached`) from recomputation, and switches UI rendering to read locks instead of write locks. It also removes the temporary demo-counter feature and updates state initialization to warm the ranking cache, whereas Side B is primarily a UI correctness and presentation fix (vote highlighting, slider/history rendering, per-group gradients) with accompanying tests.