Add SAM3 collector, dataset merger, and docs/tests for reorganized pipeline

[Copilot]

Addressed the repo reorganization into data_collection/training/testing, added reverse split tooling, and introduced a SAM3-based segmentation collector for language-grounded tracking.

• Repository layout & CLI
  • Expose data_collection/training/testing via lazy exports; add CLI commands for merge (undo splitter) and create-seg-sam3.
• Dataset utilities
  • New merge_yolo_dataset to flatten split train/val back to images/ and labels/ with collision guards.
• SAM3 language-grounded segmentation
  • New SAM3SegmentationCollector with tracking hooks, configurable patterns/template, and ontology-aware label mapping.
• Tests
  • Added pytest coverage for splitter→merger roundtrip and SAM3 collector tracking/label writing.

Example:

# Merge split YOLO dataset back to flat layout
python -m yolo_tuning.vision_tuning.cli merge --dataset-dir dataset_det

# Run SAM3 language-grounded segmentation over existing frames
python -m yolo_tuning.vision_tuning.cli create-seg-sam3 \
  --images-dir raw_frames --dataset-dir dataset_seg --prompts "a mug" "a bottle"

Original prompt

Please complete the following tasks:

1. Complete refractoring and re-organization of the repository into data_collection, training, and testing pacackages, and update readme.md and readme_CH.md to match
2. In data_collection, add merger.py which undoes the effect of splitter so that yolo format datasets can be merged back into images and labels subdirectories for further data appending
3. Create an addition segmentation data collector using SAM3's language grounded segmentation and tracking abilities.
4. Ensure there are unit tests for all of the above features

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.