Add support for transform property

[ephemer]

We now have working transforms - nested and otherwise. I haven't tested this with rotations, but since we're not using them I didn't think this was necessary. That said, most of the hard work of this implementation is done by the GPU, so there's no good reason they shouldn't work too.

What appears to not be working yet is:

• [x] Hit testing
• [x] Frame animation tests are failing
• [x] Performance optimisation (I commented it out for now, but should be easy enough to implement)

[ephemer]

@michaelknoch @rikner @cshg hitTest and absoluteOrigin are now both working as expected with transforms. https://github.com/flowkey/UIKit-SDL/pull/144/commits/31e0bd54bb8d0973a97bd1932514c4b084e0b082 is the relevant change for absoluteOrigin. Feel free to test and hopefully merge this, and then point #147 at master and merge it too if you think it's ready.

Nice to have improvements:

• [ ] absoluteOrigin should be private to UIView. Instead we should use convert(view.bounds.origin, to: window). We only use absoluteOrigin in two places, both of which are in UITouch.swift. @michaelknoch this would be a good reason to implement the UITouch.window instance property: we could then replace the absoluteOrigin call with the one above. The tests can be easily fit to use the public API as well deepestSubview.convert(deepestSubview.bounds.origin, to: rootView). I am not certain this will work, but I can't imagine how else iOS does it, and we call (the then private) absoluteOrigin anyway in our slow path for convert.
• [ ] split the five convert functions in UIView into their own file. this would be useful for cleaning up the private functions too.
• [ ] as always, add more tests. I added another view in f70e68d to ensure absoluteOrigin is working correctly. It would probably make sense to add a simpler test case for just this, now that it's clear that that is what we should have been testing for all along. (Our previous implementations passed all the tests, but broke in practice - we needed at least one extra layer between the transformed ones like this to get the implementation right)

It's probably a good idea to merge rather than squash this PR, because there's quite a bit happening and the commits are somewhat reasonably split into logical units