In fact, overall support of Trump could actually go down while the percentage of Republicans who support him goes up.
(Ie lots of people cease being Republicans, leaving only those who support Trump.)
From the what Silver was saying, I think the figures he was looking at also included an element for how much people supported Trump – you know, those surveys with "Rank from 1 to 5, where 1 is Strongly Agree, 5 is Strongly Disagree."
So if the media just take a single, headline score, the reader can't distinguish between
(scenario a) equal numbers of people scoring Trump 1 and 5
and
(scenario b) many people score Trump 4, and a much smaller number score him 1.
But these will pan out differently in an election, where each person has one equal vote.