Ltntorch

This is the third release of LTNtorch.

This version simply solves issues related to the introduction of new data types in PyTorch.

When using guarded quantification, the tensor used to mask the truth values in input has to be of type BoolTensor.

PyTorch introduced a new type, named torch.cuda.BoolTensor.

This new version of LTNtorch solves the problems that occurred when using guarded quantification on GPU since it now permits this new type.

This is the second release of LTNtorch, named v1.0.0.

In this new release, the following improvements have been made:
1. the SatAgg aggregator now accepts both torch.Tensor and LTNObject instances, instead of only LTNObject instances as the previous version. This update makes the aggregator more flexible;
2. the efficiency of the framework has been improved;
3. guarded quantification has been totally re-implemented. Now, the mask is directly passed to the aggregator and used by it to filter the grounding of the formula;
4. some problems with deep copies inside the framework have been fixed. This was the solution for the issue opened on the repository.

Initial **LTNtorch** release.

The release includes all the functionalities explained in [Logic Tensor Networks](https://arxiv.org/abs/2012.13635). In particular, the most important functionalities are:
- constants and variables;
- predicates and functions;
- connectives and quantifiers;
- diagonal and guarded quantification.