[LabelModels] The LabelModel.predict should optionally return a numpy array

[dcfidalgo]

Right now the LabelModel.predict method returns a list of records that include the predictions. This is fine if you want to log these records directly to Rubrix again for inspection, but not so good if you want to continue using the predictions for training.
Example from our tutorial:

label_model = Snorkel(weak_labels)

# get records with the predictions from the label model
records = label_model.predict()

# build a simple dataframe with text and the prediction with the highest score
df_train = pd.DataFrame([
    {"text": record.inputs["text"], "label": weak_labels.label2int[record.prediction[0][0]]}
    for record in records
])

# fit the classifier
classifier.fit(
    X=df_train.text,
    y=df_train.label
)

I would like to simplify this to:

label_model = Snorkel(weak_labels)

# get records with the predictions from the label model
records, y = label_model.predict(return_y_array=True)

# fit the classifier
classifier.fit(
    X=records.to_pandas().text,
    y=y
)

[dcfidalgo]

For multi-label predictions you would need a threshold parameter, I don't think the increase in complexity is worth this feature.
#1442 addresses part of the solution described above by returning a DatasetForTextClassification instead of a list of records.