GH-2717: Add option to ignore labels in dataset

[alanakbik]

This PR adds the option to ignore selected labels in any dataset that inherits from ColumnCorpus (i.e. nearly all datasets in Flair).

You can use the label_name_map to achieve this, by mapping all labels you don't want to learn to 'O'. The following snippet shows how to rename and ignore WNUT 17 labels so that it looks like the NER classes from CoNLL-03:

from flair.datasets import WNUT_17

# load WNUT 17 corpus with all regular labels
corpus = WNUT_17(in_memory=False)
print(corpus.make_label_dictionary('ner'))

# load WNUT 17 but rename the label 'person' to 'PER', location to 'LOC' and both 'group' and 'corporation' to ORG
corpus = WNUT_17(in_memory=False, label_name_map={
    'person': 'PER',
    'location': 'LOC',
    'group': 'ORG',
    'corporation': 'ORG',
})
print(corpus.make_label_dictionary('ner'))

# load WNUT 17 and rename like above but also ignore all 'creative-work' and 'product' entities
corpus = WNUT_17(in_memory=False, label_name_map={
    'person': 'PER',
    'location': 'LOC',
    'group': 'ORG',
    'corporation': 'ORG',
    'product': 'O',
    'creative-work': 'O', # by renaming to 'O' this tag gets ignored
})
print(corpus.make_label_dictionary('ner'))

This should print:

Dictionary with 4 tags: <unk>, PER, LOC, ORG