Improve layout abstraction to allow geometry transformations

[inclement]

Currently layouts are very simple, but barely abstracted at all from the bare positioning and sizing of views.

Ideally the layout logic should be separated from its application to views so that transforms can be applied to the geometry logic alone. In particular the idea would be make it easy to mirror or rotate layouts, avoiding needing to have a new layout for every permutation.

We'd need to do something like have the layout function act on view stand-in structs in 0-1 coordinates, then manipulate the results, then translate into real coordinates to apply to the real views.