Adds ZELDA Entity Linking dataset

[alanakbik]

Adds the ZELDA dataset for training and evaluating entity linking, introduced in our upcoming EACL paper "ZELDA: A Comprehensive Benchmark for Supervised Entity Disambiguation".

This snippet shows how to load the corpus and inspect the entity links.

from flair.datasets import ZELDA

# get Zelda corpus and print statistics
corpus = ZELDA()
print(corpus)

# get an example sentence and print
sentence = corpus.test[1]
print(sentence)

# iterate over linked entities in sentence and print each
for entity in sentence.get_labels('nel'):
    print(entity)