Fix #529: support YOLO datasets in the image detection evaluator

[gtfrans2re]

Closes #529

What changed

• Added load_image_detection_dataset() and get_image_detection_dataset_key() in
  gui/tasks/image_detection/dataset_viewer.py, following the existing
  load_image_segmentation_dataset() pattern.
• The Dataset Viewer and Evaluator tabs now share this loader, so the Evaluator can
  load YOLO datasets on its own instead of only when the Viewer had cached them first.
• The dataset type is now part of the session cache key, so switching between COCO
  and YOLO on the same folder/split no longer returns a stale dataset.
• The YOLO config is read with getvalue() instead of read(), which returns empty
  bytes on a second read of the uploaded file.

Why

As @dpascualhe noted in the issue, YOLO is supported, but the Evaluator's fallback
loader only handled COCO. Whenever the Viewer hadn't cached the dataset (e.g., the
YAML wasn't uploaded yet), users saw "Only COCO datasets are currently supported
for evaluation."

Design decisions

• The loader keeps filenames relative. model.eval() already calls
  make_fname_global() on its own copy of the dataset, and the Viewer relies on
  relative paths, so the Evaluator no longer globalizes the shared cached dataset.
• The sidebar is unchanged: YOLO stays selectable because it is supported.
• Two pre-existing lines in tests/gui/test_image_detection.py that Black would
  reformat were left untouched to keep the diff scoped to this issue.

How it was tested

• New GUI test reproducing [Bug] GUI Dataset Type selector includes YOLO but Evaluator only supports COCO — causes silent mismatch #529 (fails on master, passes with this PR).
• New unit tests for the loader: YOLO split filtering, missing YAML error, COCO
  loading, and type-aware cache keys.
• Full suite: 86 passed, 1 skipped (23 GUI tests, including 5 new ones).
• Manual check in the Streamlit app with Ultralytics coco8 (YOLO format, val split,
  4 samples).

Screenshots

YOLO dataset loaded in the Evaluator:

Same dataset in the Dataset Viewer:

[gtfrans2re]

Hi, @dpascualhe! Any plans to check this PR and merging if it's fine with you?