Build classification datasets with extract_chips

[munshkr]

We could add a --dataset-type option to extract_chips, to choose between two modes:

• classification: Each chip belongs to a class. Chips will be extracted and stored on separate directories per class.
• segmentation (default): Same as current behavior. Chips and extent masks (or boundary/distance masks) are generated.

Example:

satproc_extract_chips ./image.tif \
    --dataset-type classification \
    --labels labels.geojson \
    --classes A B \
    -o ./chips/

Output directory structure:

./chips/
    A/
        0_0.tif
        0_1.tif    
        ...
    B/
        0_0.tif
        0_1.tif    
        ...