Fix(F1Score): correct computation of F1Score in automl training

[rizoudal]

This pull request addresses issue #224.

Changes made:

• Added F1Score compution to automl training block.
• Added labels cast to prevent type errors.

Files modified:

• aucmedi/automl/block_train.py
• aucmedi/data_processing/data_generator.py

Please review the changes and let me know if there are any questions or further adjustments needed.

[muellerdo]

F1Score computation of tensorflow is buggy? Seems like shape is not really passed through from our DataGenerator/Sequence object.

Wait for fix of TF.

Possible hotfix: Setting shape ourselves with add_variable() or something??