Use cattrs to optimize response serialization

[JWCook]

Since we're already using attrs, there would be several potential benefits from using cattrs, which could be considered a pre-serialization/post-deserialization optimization step. It uses attrs definitions to recursively unstructure classes and nested data structures into dicts and builtin types, and vice versa. In relational db terms, this is similar to adding a serialization library (like Marshmallow) on top of an ORM framework (like SQLAlchemy), but without needing to define a separate model + serialization schema. There are some more details on how it works in this article.

This would make both serialization and deserialization a little more efficient, avoid some edge cases inherent with pickling (or more specifically unpickling classes), and open up the possibility for other serialization formats like JSON.

[JWCook]

Here are requests-cache's serialization modules that make use of attrs + cattrs, for reference: https://github.com/requests-cache/requests-cache/tree/main/requests_cache/serializers For this library, I would opt to keep things a bit simpler since we don't currently have a need for some of the more advanced features supported there.

I'll work on this next, whenever I have a free weekend to dedicate to it.