openai/gpt-chat-latest → A (4:1)

Side A fixes a user-visible correctness issue by generating vote URLs with `display_path()` instead of stored full URLs for left/right/pool items, matching the displayed DSL paths, and adds an exhaustive browser test that votes all 45 item pairs and verifies the resulting ranking via `GetGardenRank`. Side B mainly restructures the CLI (`forum list/show/post`), updates documentation/help text, and adjusts tests and command strings to the new interface, which is useful but is largely an API/UX reorganization rather than a core correctness improvement.