Refactoring of AutoModel logic

[alanakbik]

This PR adapts the auto-loading logic from #3011 such that the load function will distinguish between abstract and non-abstract classes:

• When called on a regular class like SequenceTagger, it will fetch/load a model as always
• When called on an abstract class like Model, it will try all non-abstract subclasses to infer the type and then load the model

This removes the need to register supported models in each class and automatically allows loading through the class hierarchy.

So the regular way for loading a model still works:

sentiment_tagger = TextClassifier.load('sentiment')

Now also these options work:

# load sentiment model through Model base class
sentiment_tagger = Model.load('sentiment')

# load sentiment model through Classifier base class
sentiment_tagger = Classifier.load('sentiment')

Since most models inherit from Classifier, you can load and run multiple different models with the same code. So, to run three different taggers for sentiment, entities and frames, do:

from flair.data import Sentence
from flair.nn import Classifier

# load three taggers to tag entities, frames and sentiment
tagger_1 = Classifier.load('ner')
tagger_2 = Classifier.load('frame')
tagger_3 = Classifier.load('sentiment')

# example sentence
sentence = Sentence('Dirk celebrated in Essen')

# predict with all three models
tagger_1.predict(sentence)
tagger_2.predict(sentence)
tagger_3.predict(sentence)

# print all predictions
for label in sentence.get_labels():
    print(label)