Create Backbone base class

[jbischof]

Pilot for how base classes could improve the maintainability of our library. The Backbone class can be easily extended to other model types and Task, Preprocessor, and Tokenizer base classes can also be added.

I've also simplified the from_preset docstring so we don't have to write a lot of custom copy for each model class. We already have all the information we need in the class docstring.

Part of #530

[mattdangerw]

Thanks! This will be a nice change. Left some comments.

[mattdangerw]

LGTM! Thanks!

[jbischof]

New from_preset docstring example below. Now the list of presets has quotes 😄

"""Instantiate BertBackbone model from preset architecture and weights.

        Args:
            preset: string. Must be one of "bert_tiny_en_uncased", "bert_small_en_uncased", "bert_medium_en_uncased", "bert_base_en_uncased", "bert_base_en", "bert_base_zh", "bert_base_multi", "bert_large_en_uncased", "bert_large_en".
            load_weights: Whether to load pre-trained weights into model.
                Defaults to `True`.

        Examples:
        ```python
        # Load architecture and weights from preset
        model = BertBackbone.from_preset("bert_base_en_uncased")

        # Load randomly initialized model from preset architecture
        model = BertBackbone.from_preset(
            "bert_base_en_uncased",
            load_weights=False
        )
        ```
"""