I suspect Cindy meant "of registered voters" not "of votes cast".
There were 153.1 million registered voters in 2008, and Obama got 69.5 million votes, thus 45.4%. So the number checks out, but that's not what the poll was asking.
Every poll selects a pool of people to ask questions from. Thus, there is no point asking people too young to vote who they would vote for. In this case, the poll's pool was "registered voters", as opposed to "likely voters" or "actual voters". They find out who is in the pool either by looking at election rolls, or asking them a question (are you registered? are you likely to vote?). Then if the person is in the pool they want, they ask a bunch of questions. In this case, one question was if they would vote for or against Trump if he's a candidate in 2020. 56% said against, something like 36% said for, and the remainder were undecided. Even if Trump picked up all the undecideds, he would not win, unless the actual voters vs registered voters skewed towards the Republicans.