Add ROCES and NCES2

[Jean-KOUAGOU]

Both ROCES and NCES2 are integrated in Ontolearn now.

• Some scripts partaining to NCES have been refactored to account for NCES2 and ROCES, e.g., the NCESTrainer class
• NCES2 and ROCES are added in tests
• NCES, NCES2, and ROCES can be run without available embeddings or pretrained models, see for example the test script, line: https://github.com/dice-group/Ontolearn/blob/add_roces_nces2/tests/test_roces.py#L38

What is left?:

Train NCES2 and ROCES and upload pretrained models to dice-files-system.

[alkidbaci]

That's quiet a large amount of changes. Overall things look good. I have a few comments on this PR:

• Add the boolean argument return_node to the method best_hypotheses where user can select to return the node or the CE (the return type should also be updated). I noticed that you save the nodes in the class argument best_predictions but in order to keep the same signature for this method throughout our learning models I recommend updating it.
• You have 2 implementations for static method generate_training_data which are very similar. You can optimize this current state by merging them into a single method and it can be moved to nces_utils.py.
• Please add docstrings for every method\class that does not inherit any docstring from a base class. This information shows in the API at our docs and it helps everyone into better understanding your code and how they can use it.

[Demirrr]

@alkidbaci please review it