The issue is quite evident on first glance. The new CNN poll out of IA uses a sample that is not even remotely plausible.
There is no way the male share of the vote will increase by 10% in relation to the female vote. That just isn't plausible. This is a gender gap that yielded an Obama victory here in 2008. There is no reason to believe Clinton faces more trouble here in 2016 than she did in 2008. This is particularly true considering how well she is doing comparatively with women than in 2008.
So, what happens when we shift the polling sample to a 2008 split? Sanders still leads with this sample, but the lead is less than half what it is with the sample CNN used. Unfortunately, CNN doesn't provide full splits and doesn't provide full cross tabs. Therefore, we are unable to make much of a judgement as to where the rest of the Sanders lead comes from when the aggregate shows that he is losing in Iowa.
I am going to opt for the third option. By adding the adjusted top-lines, it does still give Sanders a positive poll, but it gives him a lead that (while out of step with the average), is much more plausible when it comes to who will come out on caucus night.