…h as GPU clipping, scroll layers with transforms etc.

* Fix mix-blend-mode.
* Use AABB tree per render target / scroll layer.
* Change composite batching to create render target per tree layer.
* Refactor stacking context flattening to clean up code.
* Add debugging support to freelist.
* Handle render target dependencies for composite batching.
* Fix framebuffer readback during composite batching.
* Fix various warnings.
* Change render target allocation strategy (prep work for later).

(This passes tests but causes some perf regressions that will be fixed in follow up patches).