(Re)implement model cards for the training examples

[anton-l]

#748 didn't account for diffusers.hub_utils.create_model_card() (called by hub_utils.push_to_hub() which was replaced by repo.push_to_hub()) that was responsible for creating model cards like https://huggingface.co/anton-l/ddpm-ema-flowers-64 , so now they come out empty.

But since the model card generation logic is now available in huggingface_hub, we can start implementing it across all of the the training scripts (not just the unconditional example).

[patrickvonplaten]

Agree

[sayakpaul]

This has been implemented for most widely used examples.