Add functionality to split RawImage into channels; Update slice documentation and tests

[BritishWerewolf]

Some models, like U2Net, require images to be processed with all the reds, greens and blue pixels grouped together.

As a result, this PR provides the ability to do something like this.

// Get our image, then split into individual channels.
const image = await RawImage.fromURL("https://picsum.photos/600/400");

// ... snip ...
// Preprocess the image by resizing and padding to a square of 320x320
// See these PRs for functionality to do this: #971, #972

// Image pixel values are an array of [R, G, B, R, G, B,…]
// Use this to separate out each individual channel.
const [R, G, B] = image.toChannels();
// Convert from an array of Uint8Array into regular array, then flatten.
// We now have an array of [R, R, …, G, G, …, B, B, …]
const image_data = [R, G, B].map(d => Array.from(d)).flat();

// Create the tensor for this image
const tensor = new Tensor('float32', image_data, [1, image.channels, image.width, image.height]);
const model = await AutoModel.from_pretrained('u2netp');
const outputs = await model({ "input": tensor });

[xenova]

Great PR! Thanks! :) I've just applied some code suggestions to improve performance.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[xenova]

Thanks again @BritishWerewolf! 🔥 To align better with PIL's split() function, we now return new RawImage's from the function. However, if you only need the channel data, you can simply do R.data (for example)