There were two bugs in PorterDuffImageComposite.
The first one is the fact that it uses the mask input directly as alpha, missing the conversion (`1-a`). The fix is similar to c16f5744.
The second one is that all color composition formulas assume alpha premultiplied values, while the input is not premultiplied.
This change fixes both of these issue.
Previously, JoinImageWithAlpha required a batch of images to match a batch of masks. But for some use cases it's easier to provide a batch of images and a single mask.
This change automatically repeats the mask for all images in a batch.
In the same spirit, PorterDuffImageComposite will now allow a single mask for a batch of images (for both src and dst).
But also, PorterDuffImageComposite will apply the same logic to src and dst: if src contains one image, and dst is a batch it will repeat src to match dst (or the opposite).
