Allow releasing text files of tags alone without affirming ownership of images.

While there’s legal and file size issues with redistributing images, they don’t apply to the text part of training data.
Why?:
It helps users a lot to know what kind of tags were used in training (did a particular aspect of an outfit get tagged?), even if there isn’t a matching image. It’s also helpful to future LoRA makers to know how to tag images by looking at examples, and helps new LoRA makers if other people can look at their tagging style and offer feedback.