openai/gpt-chat-latest → A (3:2)

Side A makes a functional change to sibling navigation by placing each unranked sibling into its own navigation group instead of one combined group, and adds a targeted regression test verifying the new grouping behavior. Side B mainly removes a redundant zero-ratio early return and updates the test to reflect that items are still registered while zero-weight edges are skipped, which is a worthwhile cleanup but has a narrower long-term impact.