Add deserializer interfaces and document implementation of them

[JordonPhillips]

Depends on #245

This adds interfaces for shape deserializers, as well as document implementations for those interfaces. See #245 for additional context on why this approach is being taken.

Another advantage with both document serializers and deserializers available is a sort of compatibility layer with botocore:

botocore_style_input: dict[str, Any] = {"foo": "bar"}
smithy_style_input: InputShape = Document(botocore_style).as_shape(InputShape)

smithy_style_output: OutputShape = OutputShape(foo="bar")
botocore_style_output: dict[str, Any] = Document.from_shape(smithy_style_output).as_value()

There will of course be some cases where this won't work, mostly regarding cases where botocore has customized the type or name of a parameter, but for the vast majority of inputs and outputs, this should work. We could write a compatibility package that addresses those extra cases if needed.

In this way people could migrate gradually without much difficulty.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[mtdowling]

This is a test case, but in actual generated types, generating member schemas so that you can reference them directly from deserializers would avoid the overhead of hashing and getting the schema from a hashmap. Looking up schemas in the hashmap is kind of canceling out the benefit of deserializing based on the generated member index. I think you'd want those to be _ or even __ prefixed.

[JordonPhillips]

In Python, property access is usually just a dict access, so there's no real difference. I'll be experimenting with slots to speed that up, but it'll add a good bit of generated code

The base branch was changed.

[JordonPhillips]

Had to rebase