Refactor/get model allow registering of new models with function calls

[badGarnet]

This PR allows us to register new models and use them in the system:

• refactor of get_model so it relies on information stored in model_class_map and model_config_map to initialize a new model with a given model name (which is the key to both mappings)
• a new function unstructured_inference.models.base.register_new_model now allows adding new model definition to the class mapping and config mapping
• after calling register new model one can now call get_model with the new model name and get the new model

testing

New unit tests should pass

[scanny]

LGTM. One suggestion on typing conventions.

[scanny]

If you add:

from __future__ import annotations

at the top of the file you can use modern notation for union and tuple types like:

tuple[
    dict[str, type[UnstructuredModel]],
    dict[str, dict | LazyDict],
]

which does not require the imports from typing (for those types) and is consistent with how Python and IDEs display types in messages etc.

Btw, a bare dict is not a type. Here maybe it's dict[str, Any].

[badGarnet]

ah; forgot about future... I was back and forth with lint there 😅