Handle sum types in ingress encoder

[worstell]

When decoding into a type enum schema.Enum, handle using the type registry

[alecthomas]

I'm thinking maybe we should just extend our existing encoding.Marshal/encoding.Unmarshal instead, but let's discuss in the standup.

[alecthomas]

Pros:

1. omitempty in the stdlib doesn't support zero structs and thus doesn't support ftl.Optional resulting in teams using pointer fields
2. lowerCamelCase and snake_case are very common in web APIs and Go's encoder has no way of specifying those cases easily, instead requiring every field to be tagged. Very tedious. We could support that easily.

Cons:

1. Quite a lot of work.
2. We'd need to support (perhaps optionally) encoding/json.Unmarshaler/Marshaler
3. It's currently the internal encoding only, we might want to fork it instead.

This might be better as a separate open source project.