I was thinking this too at first, but this is looking at percentiles of “non-partisan” voters. I don’t see why there would be such a linear correlation on percentage of non-partisan voters.

Sure, I think the basis of your claim is absolutely valid but that should manifest simply as a random rate of non-partisans voting less for Trump. We don’t see this though, we see a very clear linear correlation (with added randomness so it wouldn’t be super obvious) that non-partisan voters voted against Trump. This shouldn’t be conflated with voting straight-Democrat.