Major refactoring of internal label logic

[alanakbik]

This PR makes a refactoring to Flair's internal label logic.

In detail:

• complex label classes like SpanLabel, RelationLabel etc. are removed in favor of a single Label class for all types of label
• each Label now has a pointer to the data point to which it belongs. This means that labels cannot be instantiated without a DataPoint object
• the Token, Span and Relation data points not inherit from _PartOfSentence, a new special DataPoint subtype. They now require a pointer to the Sentence object from which they stem. The new logic causes all labels added to a _PartOfSentence also get registered to the Sentence. So instead of previously:

sentence = Sentence("Humboldt Universität zu Berlin is located in Berlin .")

# create span
span = Span(sentence[0:4])
# make a Span-label
span_label = SpanLabel(span=span, value='University')
# add Span-label to sentence
sentence.add_complex_label(typename='ner',  label=span_label)

you can now just get a span from the sentence and add a label to it directly. It will get registered on the sentence as well.

# get Span
span =  sentence[0:4]
# add label (will automatically register to Sentence as well)
span.add_label("ner", "Organization")

• this in turn simplifies the signature of the forward_pass method of DefaultClassifier to return 3 instead of 4 arguments (Sentences no longer needed). It also does away with the unintuitive spawn logic we no longer need.
• a number of fields have been added or moved up to the DataPoint class, for convenience, including properties to get start_position and end_position of datapoints, their text, their tag and score (if they have only one tag) and an unlabeled_identifier
• a number of methods like get_tag and add_tag have been removed from Token in favor of the get_label and add_label method of the parent DataPoint class
• the get_spans method of Sentence is back, and a similar get_relations method added
• the Tokenizer classes no longer return lists of Token, rather lists of strings that the Sentence object converts to tokens, centralizing the offset and whitespace_after detection in one place
• many unit tests added