Optimize Split Transparency output for subsequent operations

[RunDevelopment]

While optimizing resize, I learned that creating an RGB image by returning the first 3 channels of an RGBA image can cause very significant slow downs in subsequent operations. This is because iterating over array slices is a lot slower than iterating over contiguous memory. Slices also might require a copy to make them contiguous (some image operations are implemented in such a way that they require contiguous memory).

I also want to say that we generally don't need to worry about slices being bad for performance. They are only bad if they can be trivially avoided, like in Split Transparency. For RGB images, this node was creating a new RGBA image and then returned 2 slices to that RGBA image. So I added some code to return the RGB image and new image instead. This not only improves performance for subsequent operations, it also allocates less memory.