Blending Learning to Rank and Dense Representations for Efficient and Effective Cascades