When we have a scaling transform applied to a leaf layer (and we don't think the transform is likely to be animated), we can apply the scaling component of the transform into the contentsScale of the layer. This will give us better quality results for upscaled content, and less memory usage for downscaled content.

There's a potential for performance regressions here, so we want to try to limit it (for now!) to cases where we're confident it's going to be an improvement.

<rdar://89515405>