models: add support for Hugging Face model hub

[stefan-it]

Hi,

this PR allows to use trained sequence tagging from the Hugging Face model hub 🤗

It is only needed to specify the namespace, like stefan-it/flair-ner-conll03 and a model will be downloaded (incl. caching) that can be used in Flair!

Naming convention is, that the model file on the model hub is named model.bin.

The following code snippet shows how to use this new feature:

from flair.data import Sentence
from flair.models import SequenceTagger
tagger: SequenceTagger = SequenceTagger.load("stefan-it/flair-ner-conll03")

It is also possible to pass a specific commit or branch name via @-syntax: stefan-it/flair-ner-conll03@main.

On the Hugging Face model hub itself, it is also possible to add a README file, that gives more details about a trained model:

https://huggingface.co/stefan-it/flair-ner-conll03

Many thanks to the Hugging Face team for the awesome model hub ❤️

[stefan-it]

As discussed with @alanakbik , I changed the cache dir (from huggingface cache dir) to the flair one 🤗

In [2]: tagger: SequenceTagger = SequenceTagger.load("stefan-it/flair-ner-conll03")                                                           
Downloading: 100%|█████████████████████████████████████████████████████████████████████████████████████████| 432M/432M [00:39<00:00, 11.1MB/s]
2020-11-23 12:49:20,577 loading file /home/stefan/.flair/36e1432823a81d0a0e298b5847412de75d780104bc5e2edb2ddd61b63b8f7436.941c7c30b38aef8d8a4eb5c1b6dd7fe8583ff723fef457382589ad6a4e859cfc

In [3]: tagger: SequenceTagger = SequenceTagger.load("stefan-it/flair-ner-conll03")                                                           
2020-11-23 12:49:26,667 loading file /home/stefan/.flair/36e1432823a81d0a0e298b5847412de75d780104bc5e2edb2ddd61b63b8f7436.941c7c30b38aef8d8a4eb5c1b6dd7fe8583ff723fef457382589ad6a4e859cfc

[alanakbik]

Thanks! Can you also pin the transformers version between 3.5.0 and 3.5.1 so that this always works?