Fix the image of the architecture overview in our documentation

[akihironitta]

If I understand it correctly, the image below doesn't reflect our current structure precisely because the output of FeatureEncoder should be a concatenated tensor. However, the image illustrates that the concatenation is applied outside the FeatureEncoder.

[yiweny]

The concatenation is done by the StypeWiseFeatureEncoder after applying each StypeEncoder to each stype. Since we are adding the ColumnWiseFeatureEncoder soon, which means that columns of the same stype might not even be concatenated together, we can defer this task till then.

[akihironitta]

@yiweny Thanks, I agree that this is not urgent.

Since we are adding the ColumnWiseFeatureEncoder soon, which means that columns of the same stype might not even be concatenated together

Still, it might be better to have a consistent interface to make other parts of PyF models FeatureEncoder-agnostic, which contributes to the modular design of PyF.