fix has unknown label is not always initialized

[helpmefindaname]

fixes #3367

[alanakbik]

@helpmefindaname thanks for fixing this! The error seems to occur because there is no valid data point to be predicted on in the test set, and therefore the has_unknown_label is never initialized.

When testing with the example script below, this PR does indeed fix this error. However, it throws another error later when the evaluate code tries to access the classification_report_dict["accuracy"] field - since there is no valid data point to make a prediction, no accuracy can be computed.

This error can be reproduced with:

import flair
from flair.data import Corpus
from flair.datasets import ColumnCorpus, WNUT_17
from flair.embeddings import TransformerWordEmbeddings
from flair.models import SpanClassifier
from flair.trainers import ModelTrainer

flair.set_seed(123)

corpus: Corpus = WNUT_17().downsample(0.01)

label_dictionary = corpus.make_label_dictionary("ner", add_unk=True)

embeddings = TransformerWordEmbeddings(
    model="distilbert-base-uncased", layers="-1", subtoken_pooling="first", fine_tune=True, use_context=True
)

tagger = SpanClassifier(embeddings=embeddings, label_dictionary=label_dictionary)

trainer = ModelTrainer(tagger, corpus)

trainer.fine_tune(
    "./results/",
    max_epochs=1,
)

[alanakbik]

The second error is another issue, merging this now and opening a separate PR for the above error.