feat: add LibreYOLO as detection model backend

[IIIllllIlIlllII]

Summary

Adds LibreYOLO as a new detection model backend, as requested in #1397.

LibreYOLO is an MIT-licensed computer vision library with an API compatible with Ultralytics (same Results, Boxes, Masks, OBB interfaces). The implementation inherits from UltralyticsDetectionModel with four overrides to adapt for LibreYOLO's differences:

• check_dependencies() — requires libreyolo instead of ultralytics
• load_model() — uses LibreYOLO() factory (device is passed at construction, not via .to())
• perform_batch_inference() — removes the cfg kwarg that LibreYOLO doesn't support
• _extract_predictions() — imports Masks from libreyolo instead of ultralytics

Usage

from sahi import AutoDetectionModel, get_sliced_prediction

detection_model = AutoDetectionModel.from_pretrained(
    model_type="libreyolo",
    model_path="LibreYOLO9t.pt",
    confidence_threshold=0.3,
    device="cpu",
)

result = get_sliced_prediction(
    "image.jpg",
    detection_model,
    slice_height=640,
    slice_width=640,
)

Changes

File	Description
sahi/models/libreyolo.py	New — model class inheriting from UltralyticsDetectionModel
sahi/auto_model.py	+1 line — register "libreyolo" model type
sahi/utils/libreyolo.py	New — test constants and model download helper
tests/test_libreyolo_model.py	New — 5 tests (load, set_model, inference, convert, auto_model)

Tests

All 5 tests pass:

tests/test_libreyolo_model.py::test_load_model PASSED
tests/test_libreyolo_model.py::test_set_model PASSED
tests/test_libreyolo_model.py::test_perform_inference PASSED
tests/test_libreyolo_model.py::test_convert_original_predictions PASSED
tests/test_libreyolo_model.py::test_auto_model_type PASSED

Why LibreYOLO?

LibreYOLO provides MIT-licensed YOLO inference (vs Ultralytics' AGPL), which matters for users building commercial or closed-source applications. It supports YOLOv9, RF-DETR, YOLOX, and other architectures behind a single LibreYOLO() factory.