openai/gpt-5.3-chat → A (3:1)

A adds a concrete behavioral improvement by filtering out stickied/pinned Reddit posts in `parse_children`, with a dedicated helper and test ensuring only real posts are imported. B mainly removes a supposedly dead zero-ratio guard in `apply_vote` and adjusts tests, which is a cleanup with minor semantic change but less clear user-facing impact.