Presidential elections always have more democrats voting because many blacks and Hispanics don't vote in off year elections. 2012 was the first year that blacks voted in a greater percentage than whites.
The other thing is that polls are as of today. A poll today is not necessarily reflective of November, it is if the election is held today. I have to assume that the democrat +8 percent will vary. If it is a bad election for the democrats, their turnout will be depressed and republicans could very well be competitive in turnout.
Yes, and that's why the party identification is important. If the ID breakdown is significantly different than recent comparable elections, you have to at least consider the effect on the outcome of the poll.
("Push polls" become obvious when you consider this metric. Some pollers conduct these "polls" to reinforce a pre-determined outcome, and make their candidate look good.)
And, you have to consider whether this is a registered or likely voter sample. At this point, registered vs. likely is probably meaningless, because I don't know how one can reliably measure that 5-6 months early. But, it can be a useful measurement of voter enthusiasm.
But, since a random phone poll only uses people that accept the call, it tends to over-sample people that are at home and answer the phone. This almost always oversamples Democrats.