The view target box isn't a good log of view position

[inclement]

The current target box information in each view stores its expected bounds, but this doesn't work well for floating views.

We should probably instead/additionally store the target view box in output-relative coords (i.e. 0-1 on each axis), which can then be used to reposition/resize it appropriately when the output is changed.

This needs more thought, but right now floating views are probably not treated correctly on workspace rearrange with multiple outputs.