Change LoRA trainer's question on rights ownership to allow public domain training data.

Currently the LoRA someone who trains a LoRA on exclusively public domain or on a mixture of public domain images and images they own the training data technically can't automatically release the training data, as they don't actually own the rights (nobody does) to public domain content. Would help historical styles, a lot of things that were documented by various governments (especially the US), and some other stuff.

This is a simple wording tweak. A slightly more complicated but more complete solution would be to also support Creative Commons with support for share alike (release the model as such), and/or attribution (add a text file where each image can be attributed to the training data).