~anthropic/claude-sonnet-latest → B (6:4)

Side B fixes a real, user-visible multi-user bug (shared sentinel delegate blocking subsequent human votes) with a targeted change, new integration test, and updated docs/types, delivering concrete correctness value. Side A is a large mechanical rename (CanonicalItemUrl -> ItemId) plus deletion of a planning doc; while structurally reasonable, it's mostly churn/refactor with no new capability or bugfix and carries risk (fallback opaque() bypassing normalization) without demonstrated necessity beyond the refactor's own stated goals.