Performance Optimization of Augmentations

[jakubMitura14]

This task aims to improve the runtime performance of data augmentation routines. The focus is on adapting augmentations that rely on interpolation (e.g., elastic deformation, low-resolution simulation) to GPU execution using the CUDA.jl-based backend developed after the last GSoC.
Success when:
Augmentations execute with lower memory overhead.
Selected augmentations (e.g., scaling, low-resolution) run on GPU.
Benchmarks are provided comparing new and old implementations.