GH-243: dataset downloader

[alanakbik]

addresses #243

• DataFetcher can now download universal dependencies corpora for 30 languages, WikiNER corpora for 8 languages and some CoNLL tasks.
• If there is only a training data file, now samples both dev and test data (Train NER for Swedish #3)
• rename all fetch_* methods to load_* methods

So now you can load a dataset like this:

from flair.data_fetcher import NLPTaskDataFetcher, NLPTask

# load one corpus
corpus = NLPTaskDataFetcher.load_corpus(NLPTask.UD_ENGLISH)
print(corpus)

# load a MultiCorpus of two UD corpora
corpus = NLPTaskDataFetcher.load_corpora([NLPTask.UD_ENGLISH, NLPTask.UD_GERMAN])
print(corpus)

You no longer need to download the UD corpus yourself. The method will check if it is there and if not download the corpus.