Extend Ultralytics Backend with Additional Vision Tasks

[kprokofi]

Description
Expand the Ultralytics backend to support additional computer vision tasks beyond the current implementation.

Goals

• Add multiclass classification support
• Add multilabel classification support
• Add semantic segmentation support
• Integrate preprocessing and augmentation pipelines for all tasks
• Maintain compatibility with existing training APIs

Acceptance Criteria

• All three tasks are supported end-to-end
• Dataset validation works for each task
• Export supported to both OpenVINO and ONNX formats
• Training and inference APIs are available
• Documentation is updated
• Automated tests cover new functionality