v3.2.1: Fix bug in protocol "segment": now the GPU selection work as expected.
v3.2.0:
 Users:
   - Deepfinder template added
   - Segment and cluster protocols semi-streamified.
   - Prevent viewers from locking the screen focus.
 Developers:
   - Replace the setup.py by the corresponding pyproject.toml.
v3.1.0:
    - Target generation and cluster protocols can be now parallelized using threads.
    - The training and validation datasets can be both introduced manually or generated from the training dataset
      specifying a fraction of data.
    - Fixes in the tomogram/tomomask assignment in the segment and cluster protocols.
    - Calling the DeepFinder viewer now offers an auxiliar window to previously select the tomogram to be displayed.
    - Add a viewer for the learning curves generated in a training.
    - The background class in now considered internally by Scipion when importing a model.
    - Critical warning about the FOV value added to the target generation protocol.
    - Beta status removed.
    - Annotation protocol now allows to generate the output even though not all the tomograms have been annotated.
      IMPORTANT: until some requested changes in the source code are carried out, the tomograms can only be annotated
      once. Otherwise, its coordinates are lost.


