feat: add YOLOv8 COCO subset evaluation demo

[hyeokjun32]

Summary

• Added a small YOLOv8 COCO-style subset evaluation report generated from local image/label pairs.
• Connected the subset evaluation summary to Local Studio Load Demo Evidence.
• Documented the subset metrics in portfolio docs without committing raw images.

Validation

• poetry run python3 -m pytest tests/test_studio_routes.py tests/test_validation_demo_report.py tests/test_validation_contracts.py tests/test_evaluate_detection.py -q
• poetry run python3 -m pytest -q
• PYTHONPATH=/Users/GwonHyeokJun/Documents/GitHub/InferEdgeLab python3 -m inferedgelab.cli evaluate-detection /Users/GwonHyeokJun/Documents/GitHub/edgebench/models/yolov8n.onnx --engine onnxruntime --image-dir /Users/GwonHyeokJun/Desktop/my_demo/images --coco-annotations examples/validation_demo/subset/yolov8_coco_subset_annotations.json --preset yolov8_coco --model-contract examples/validation_demo/yolov8_coco_model_contract.json --num-classes 80 --precision fp32 --report-json examples/validation_demo/subset/yolov8_coco_subset_evaluation.json --report-md examples/validation_demo/subset/yolov8_coco_subset_evaluation.md --report-html examples/validation_demo/subset/yolov8_coco_subset_evaluation.html --no-save-structured-result

Impact

• Adds real subset accuracy evidence to the contract/preset validation workflow.
• Lets Local Studio show both latency demo evidence and YOLOv8 COCO subset evaluation summary.
• Keeps raw image data out of the repo while preserving annotation/report evidence.

Notes

• This is a small subset demo, not a full COCO benchmark.
• No DB, queue, auth, upload, or production SaaS behavior was added.