Add wrapper for ultralytics yolo for instance segmentation task

[kprokofi]

Summary

Add YOLOSeg — a YOLO instance-segmentation model wrapper. It is needed for YOLO26 integration to Geti

Changes

• src/model_api/models/yolo_seg.py — New YOLOSeg(DetectionModel) class (model = "YOLO-seg"):
  • init validates 1 input / 2 outputs, identifies detection (rank-3) and prototype (rank-4) tensors
  • parameters() inherits full NMS chain from DetectionModel, sets YOLO-specific defaults
  • postprocess() parses detection output → confidence filter → NMS → mask decoding (coefficients @ prototypes → sigmoid → crop → resize) → coordinate scaling → returns InstanceSegmentationResult
• tests/unit/models/test_yolo_seg_model.py — 8 unit tests covering init validation, parameters, postprocessing (no detections, with detection, multiple detections), and crop_mask

[kprokofi]

Verified in application:

[kprokofi]

Not sure about serving_api-tests. "docker: error pulling image configuration: download failed after attempts=6: dial tcp 54.185.253.63:443: connect: connection refused". Probably not related to my changes

[AlexanderBarabanov]

The issue has been fixed in #595 @kprokofi