Cache images to make extractions with data augmentation quicker

Sometimes downloading the full size images can take a long time. So it's wasteful to download them again when using a different extraction mode or some other data augmentation.

Instead the images could be saved in ~/.cache/kaldi_data_generator/images/ with the path from the IIIF url.

Then it could be loaded in self.get_image if it exists.

Could also save the requests ListTranscriptions, but saving the images should already give a big speed boost.