Dataset from dicts

[tholor]

As requested in #85, we can add an option to let the DataSilo load data from dicts instead of automatic loading from files.

Exemplary Usage:

processor = TextClassificationProcessor(tokenizer=tokenizer,
                                        max_seq_len=8,
                                        data_dir=None,
                                        train_filename=None,
                                        label_list=["OTHER", "OFFENSE"],
                                        metric="f1_macro",
                                        dev_filename=None,
                                        test_filename=None,
                                        dev_split=0.0,
                                        label_column_name=None)

data_silo = DataSilo(
    processor=processor,
    batch_size=batch_size,
    automatic_loading=False)

basic_texts = [
    {"text": "Martin Müller spielt Handball in Berlin.", "text_classification_label": "OTHER"},
    {"text": "Schartau sagte dem Tagesspiegel, dass Fischer ein Idiot sei.", "text_classification_label": "OFFENSE"}
]
data_silo._load_data(train_dicts=basic_texts)