Show space used by auto captioning models and allow to delete some of them

[ChristianMayer]

Trying an auto captioner downloads it automatically. This is a great feature!

But it comes with a few draw backs:

• selecting a model I don't know that it will download multi digit gigabytes of data once activated.
• once a download is started I have no way to stop it - especially bad as that's the first time I'm learing how much data will be downloaded now.
• it's saved somewhere in the system so it's not easy to figure out how much space it's occupying - and the models are huge (doing a full filesystem search I could figure out they are stored at ~/.cache/huggingface/

So I want to suggest a few enhancements:

• give a hint about the size that will be downloaded for a model so that uses with limited space will not select it
• add a button to stop a started download ("cancel auto captioning" might be reused for that)
• show for all downloaded models the size they are taking on the hdd
• offer an option to delete a model from the hdd