4.2.4: deploy test
4.2.3: Predict protocol:
    - Add fail tolerance, loggers and improve the summary.
    - Default input expects now the odd/even tomograms linked to the tomogram object.
4.2.2: Add extra validation to the training protocol.
4.2.1: Avoiding "which" when launching cryocare commands
4.2.0:
  Users:
   - Training protocol admits multiple GPUs.
   - Predict protocol is parallelized in GPU and semistreamified.
4.1.0:
  Users:
    - Binary version installed is now 0.3.0 (https://github.com/juglab/cryoCARE_pip/releases/tag/v0.3.0)
    - We now parse GPU IDs to accept both comma-separated and space-separated specification (Thanks,
      Ricardo Righetto).

4.0.0:
 Users:
  - Prepare data protocol was removed, and now it's a step of the training protocol.
  - Hotfix: predict protocol was not setting the tilt series id to the generated tomograms.
 Developers:
  - Even/Odd present in metadata or not logic was inverted along the training and predict protocols.
  - Add a base protocol to manage these two entryways to the training and predict protocols.
  - Tests refactored using the Test Centralization Layer from scipion-em-tomo.
3.2.1: Re-deploy to pypi
3.2:
    - Use of odd even simplified and associated to the tomograms
    - generate data training removed
3.1.1:
    - Source code version 0.2.2.
    - Updating the installation command: Taking into account cuda 10 and 11 installation
3.1.0:
    - Tensorflow 2 version. Source code version 0.2.1.
