Implement YOLOE open-vocabulary instance segmentation model

[rycerzes]

Support YOLOE open-vocabulary instance segmentation models, including both text/visual prompt and prompt-free variants. This PR adds support for flexible segmentation modes via text prompts, visual prompts (bounding boxes), or prompt-free operation. The models are registered but commented out by default in the main models configuration.

YOLOE Segmentation Model Implementation:

• Added a new YOLOESegmentation model class in app/models/yoloe.py supporting text prompt, visual prompt, and prompt-free segmentation modes, with detailed inference logic, mask-to-polygon conversion, and post-processing to output shapes.

Model Configuration Files:

• Added configuration files for the following YOLOE segmentation variants:
  • yoloe_26s_seg.yaml: Text/visual prompt mode for the 26S variant.
  • yoloe_26s_seg_pf.yaml: Prompt-free mode for the 26S variant.
  • yoloe_26l_seg.yaml: Text/visual prompt mode for the 26L variant.

Model Registration:

• Registered multiple YOLOE variants (including 26n/s/m/l/x and v8 s/m/l, both prompt and prompt-free) with the @register_model decorator in yoloe.py, making them available for use in the platform.

Model Availability:

• Updated configs/models.yaml to include (commented out) entries for the new YOLOE segmentation models, allowing easy activation as needed.