Lots of pollsters do this. There's nothing wrong with that approach. Let the party ID float and you could be wrong. Try to predict the party ID and you could still be wrong. Rasmussen takes something of a hybrid approach between the two. And he could still be wrong.
Disagree, esp. in this election where you could more accurately balance party ID based on predicted outcome than simply letting it float. Clearly Republicans and independents are going to be more likely to turn out. The hybrid approach would be going off of actual registration numbers which is still more accurate than just letting it float.