This PR refactors the Token and Sentence positional properties inherited from the DataPoint to replace the self.start_pos and self.end_pos attributes.

For the Sentence, both variants of accessing positional information behaved differently, resulting in inconsistent results where they appropriately should have been aliases. These are the inconsistencies:

1. Initializing the Sentence with str, i.e. untokenized input.

   • We allow the user to set an offset start_position in the init but this is not respected in the start_position property. It always returns zero. The inconsistency is also that in start_pos the start position offset is included.
     Suggestion: Only use and expose the property inherited from the DataPoint. Having multiple attributes doing the same thing with different names may get confusing.

2. Initializing the Sentence with List[str], i.e. pre-tokenized input.

   • Same concern as in (1)
   • Added to (1), end_pos and end_position actually do not behave the same.

     from flair.data import Sentence
     s = Sentence(['This', 'is', 'an', 'example', '.'])
     print(s.end_position)  # Prints 17 -> Corresponding to the character-level end position
     print(s.end_pos)  # Prints 5 -> Corresponding to the token-level end position

     Suggestion: Always use the character-level end position since the token-level end position is accessible with len(s).
   • The token start and end positions were incorrect.

     from flair.data import Sentence
     sentence = Sentence(["This", "is", "a", "sentence", "."])
     print([(token.start_position, token.end_position) for token in sentence])  
     # Prints [(0, 4), (5, 7), (7, 8), (8, 16), (16, 17)] but expected [(0, 4), (5, 7), (8, 9), (10, 18), (19, 20)]

     Suggestion: Do not use two separate methods to construct the tokens. Instead, convert the case of initializing the Sentence with List[str] to the case of initializing the str.

Please see the commits as isolated corresponding to the suggestions.