Improving the dataset preprocessing

[plonerma]

We discussed improving how the dataset is preprocessed. I added an test-case to ensure the original dataset is not being changed (but instead a changed copy is returned) and started improving some minor points.

There are some open points that should be addressed in the future:

• The DataCleaner should be EITHER stateless (i.e. is configured in the init and preprocessing dataset does not change any attributes) OR statefull (i.e. it should only be invoke once with a dataset).
• We need to change the order of the processing steps, as at the moment string-labels do not work.

[lukasgarbas]

Did some refactoring. Datacleaner doesn't change its instance attributes or the original dataset. prepare_dataset(dataset) now returns texts, labels, and the task category.

Changed the order of data preprocessing steps. String labels are now converted to integers when creating the label map.