[FEATURE] Create whole image dataset

[LucaRom]

This feature involves creating a dataset that supports extracting samples using a moving window approach on a whole image stored locally or remotely. The goal is to provide a straightforward mechanism for handling full-resolution images or data inputs. This contrast with using pre-tiled images stored locally.

Key requirements:

• Implement a mechanism for extracting patches from the dataset using a configurable window size and step.
• Ensure support for overlapping and non-overlapping windows.
• The dataset should be compatible with PyTorch DataLoader and integrate smoothly into the PyTorch Lightning LightningDataModule.
• Optimize the process for both memory usage and performance, especially for large datasets.
• Adapt the dataset to support STAC items