Rank and select document chunks by relevance to fit within a token budget. Returns selected/excluded chunks with scores.