Fixes #611

Problem

Calling batch_image_classification(data_path=...) on any TIMM-based or ResNet-based classifier raises a TypeError because pw_data.ImageFolder does not accept a path_head keyword argument:

# Both timm_base and resnet_base had this:
dataset = pw_data.ImageFolder(
    data_path,
    transform=self.transform,
    path_head='.'        # ← ImageFolder doesn't accept this
)

Additionally, ImageFolder is the abstract base class whose __getitem__ returns None, so even if the kwarg issue were fixed, the dataloader would fail to unpack (img, img_path).

Fix

• Replace pw_data.ImageFolder with pw_data.ClassificationImageFolder — the correct subclass that implements __getitem__ returning (img, img_path).
• Remove the invalid path_head='.' argument (ClassificationImageFolder builds full paths via os.walk, so path_head is unnecessary).
• Add ClassificationImageFolder to __all__ in datasets.py for export consistency.

Files changed

Safety

• The det_results code path (using DetectionCrops) is untouched — path_head is valid there.
• No other file in the repo references pw_data.ImageFolder.
• All detectors already use DetectionImageFolder correctly.
• The broken code path could never have worked, so this is purely additive — no existing working behavior changes.