Integrations of modifications made for GLGExplainer

[steveazzolin]

I'm integrating the modifications made to the package for the development of GLGExplainer, namely:

• remove_attention parameter in EntropyLayer in order to leave the concept vector unmodified during the forward pass
• function test_explanations(...) for computing the accuracy of formulas as a classifier, i.e., considering multiple activations of formulas as an error
• simplify parameter in explain_class(...) for controlling whether to apply formula simplifications or return a dump of all formulas