Function load_supervised_image_folder_dataset 

pub fn load_supervised_image_folder_dataset<P: AsRef<Path>>(
    root: P,
    config: &SupervisedImageFolderConfig,
) -> Result<SupervisedDataset, ModelError>

Loads supervised training samples from an image-folder classification tree.

Directory contract:

• direct child directories under root are class buckets,
• class index is assigned by deterministic lexicographic directory ordering,
• supported image extensions are jpg, jpeg, and png (case-insensitive),
• non-image files are ignored.

Runtime mapping:

• inputs shape: [N, output_height, output_width, 3],
• targets shape depends on config.target_mode():
  • ClassIndex: [N, 1] with scalar class ids,
  • OneHot: [N, class_count] one-hot vectors.