[Feature] option to split dataset by defect type

[rasa-pa]

📄 Description

In our projects a single dataset often contains images captured under different lighting setups or different viewpoints, as well as multiple defect types with very different characteristics. Training all defect types together sometimes leads to poor model performance, especially when one defect type is overly frequent, extremely small or visually very different from the others.

To address this currently we manually reorganise data outside GETi.

🎯 Objective

A feature that lets users split an existing dataset into multiple datasets by defect class labels. This will let us remove dominant or problematic defect types that degrade overall performance, and isolate rare or small defects into their own model workflows.

For us this functionality would support a cleaner dataset management and improve training outcomes for complex use cases.

[piotrgrubicki]

Relates to #1553