Expanded computer vision models support (new models and apis)

[yvolokitin]

Add support for new computer vision models and extend backend capabilities by:

• integrating the EdgeCrafter model into the Lightning training engine
• expanding the Ultralytics backend with multiclass, multilabel classification, and semantic segmentation
• unifying training, validation, inference, and export APIs across backends
• aligning configuration, preprocessing, and augmentation pipelines
• ensuring support for model export (ONNX, OpenVINO) and checkpoint handling
• improving dependency management (e.g., resolving external library forks)
• providing documentation, examples, and test coverage for all new functionality