recreate to_dict and add relations

[helpmefindaname]

Closes #3265

partial revert of decd3a6

This cleans up the to_dict() iterface, giving options to get all labels, entities and relations of a sentence

[turian]

There are a few things I'd love to see in this PR, which would make me run it instead of main.

1. You seem to be dropping labels from the pos tagger by default. Why?
2. I'd really like the token start_position and end_positions in the dict, so I can recover the original text from a list of tokens. (token.text would be nice but unnecessary)

Try this code:

from flair.data import Sentence
from flair.models import SequenceTagger

# load tagger
tagger = SequenceTagger.load("flair/pos-english-fast")

# make example sentence
sentence = Sentence("I love Berlin.")

# predict NER tags
tagger.predict(sentence)

print(sentence.to_dict())

with your branch I get:

{'text': 'I love Berlin.', 'labels': [], 'entities': [], 'relations': []}

with main I get:

{'text': 'I love Berlin.', 'all labels': [{'value': 'PRP', 'confidence': 0.9999862909317017}, {'value': 'VBP', 'confidence': 0.999794065952301}, {'value': 'NNP', 'confidence': 0.9999107122421265}, {'value': '.', 'confidence': 0.9998101592063904}]}

Here is a little workaround that works with main:

from flair.data import Sentence
from flair.models import SequenceTagger

# load tagger
tagger = SequenceTagger.load("flair/pos-english-fast")

# make example sentence
sentence = Sentence("I love Berlin.")

# predict NER tags
tagger.predict(sentence)

token_list = []
for token in sentence:
    token_list.append((token.start_position, token.end_position))

sentence_dict = sentence.to_dict()
sentence_dict["token positions"] = token_list
assert len(sentence_dict["token positions"]) == len(sentence_dict["all labels"])
print(sentence_dict)

Which gives:

{'text': 'I love Berlin.', 'all labels': [{'value': 'PRP', 'confidence': 0.9999862909317017}, {'value': 'VBP', 'confidence': 0.999794065952301}, {'value': 'NNP', 'confidence': 0.9999107122421265}, {'value': '.', 'confidence': 0.9998101592063904}], 'token positions': [(0, 1), (2, 6), (7, 13), (13, 14)]}

That's what I really want!

[alanakbik]

@helpmefindaname thanks for adding this!