openai/gpt-chat-latest → A (4:1)

Side A makes a substantial architectural improvement by introducing a settlement worker that batches vote persistence and ranking recomputation, adds cached ranking reads (`ranked_items_cached`), warms the cache at startup, and switches HTTP rendering from write locks to read locks. It also removes the temporary demo counter and related event/UI code, whereas Side B is primarily a URL refactor that centralizes private-room route encoding/decoding and adds URL normalization helpers, which is useful but narrower in long-term impact.