Yep! That's exactly as I would expect. If you exactly fill the sensor, and then rotate the image, then the corners would not have physical pixels anymore to record them. So to avoid having black corners where there's literally no data, it zooms in just enough to put the physical edge of the sensor on the edge of the screen again.
If you start with it already zoomed-in, then I would bet it doesn't do that, until you rotate enough that it has to. Or, the software could be made simpler by always adding that amount of zoom regardless.
I don't know which way Apple went with *that* choice, but that's what it's doing. It's fitting a rotated rectangle entirely inside the original one, which requires it to be smaller. Then that smaller rectangle is expanded back to fill the screen, which results in a zoom-in.