Bridging the Divide: Reconsidering Softmax and Linear Attention