Audit our `keras.models` class docstrings after base class additions

[mattdangerw]

With the addition of base classes for tokenizers, preprocessors, backbones, and tasks, we are changing how we document our classes in our docstring and on keras.io.

The from_preset() docstring will now be minimal, and the class level docstring should contain code examples with from_preset.

We need to audit all our class level docstrings, and rewrite our code examples so we lead with the simplest usage case (from_preset). And then dig into the more complex cases (bring your own config + data).