openai/gpt-chat-latest → A (4:1)

Side A changes the ranking color logic from list-position-based gradients to score-based min–max normalization within each group, introduces a dedicated `score_gradient_t` helper, updates callers, and adds targeted tests covering normalization behavior and edge cases such as tied scores. Side B fixes a real configuration bug by defining `GITHUB_API_BASE_URL` with a default, but it is a small missing-variable fix with narrower impact than the broader, tested behavioral improvement in Side A.