Docs for generated code visibility best practice

[kriswuollett]

I believe most of the protobuf code generators follow the published best practices published on https://protobuf.dev/. Some examples are:

• generated code should have "hazzers", not optional types
• enums should always have an unknown for the 0 value

That's great. However, I somewhat recall a "Readability" issue in general that protos should not to be used for types within core application logic directly, i.e. protos should ideally be used for serialization only? I think such guidance is missing from the docs, and that could have the impact of unnecessary library/semver coupling showing up in open source libraries as more developers adopt gRPC/(Protobuf/FlatBuffer/etc.)? If included in the docs it could also help with the backward/forward compatibility mindset.