Add LoRA fine-tuning script
Adds fine_tuning/train_lora.py, a single-GPU LoRA trainer for editing and text-to-image on BriaFibo2Pipeline, with fine_tuning/README.md covering install, data format, training and loading the LoRA. Tested on an H100 against this repo with diffusers commit 6894a7e8c.
Card: replaces the Fine-tuning code: [TBD] line with a link to fine_tuning/.
Merged: thanks! Before launch, please port the trainer to the current diffusers PR head (fonts=, the new prompt parsing / BriaFibo2ImageProcessor); it pins 6894a7e8c, whose private methods are gone at the PR head.
FYI: multi-reference is capped at 4 images across the release, so train_lora.py now accepts 1-4 sources per row (check, error message, docstring and cache loader) and the guide says 1 to 4. Please keep that when you port the trainer to the current diffusers PR.