Creating a TextDataset with None items results in TypeError during SetFit training

[chschroeder]

Feature description

Creating a TextDataset that contains None item should be prevented.

Motivation

If you create such a dataset, which does not make sense but is currently possible, during SetFit training you end up with a strange error:

<...>
    encodings = self._tokenizer.encode_batch(
TypeError: TextEncodeInput must be Union[TextInputSequence, Tuple[InputSequence, InputSequence]]

Not really a bug, but we can check for this and provide an error that is more helpful.

Additional comments

Reported by @eisioriginal