fixes #13557

The failure was

The case was these two elements: (zero width target and passing target)

In taptargets we take each target (eg zero width target) as a rect. then we place a hypothetical 48x48 finger at the center of the rect. Then we see how many pixels of overlap it has with the adjacent target (passing target)

Test said it should be 720 pixels of overlap area, but on a high dpi screen it was 696. They're both right.

In hidpi you can have fractional CSS pixels.. so.. a lot of things are 18.5px tall. And that ends up with a very similar but slightly off result in these rect overlap calculations.

Fixing it with a font-size on the body was a guess. I never understood this whole thing but it probably is related.

The discrepancy above points to a problem with the implementation of emulation. I can repro this difference while emulating DPR=2 on my dpi1 screen and emulating DPR=2 on my dpi2 screen. One says 18px tall, the other 18.5px tall. Thats not supposed to happen. That said, this subtlety probably isn't causing us large problems, so I'm not concerned.