Improve typing

[TobiasBales]

This bumps the # typed sigil to strict and runs srb tc in test cases.

This means some adjustments to the generated code (or quite a lot).
I would like to get initial agreement on the direction and then would do some clean up (there are definitely some simplifications possible and required).

Some things/questions/observations that come to mind:

• The discrimator field for oneof values can be nilable (in generated code), even for required fields
• Enums in one of fields are currently broken, the actual implementation. This was surfaced by sorbet, will fix this as a follow up PR
• Optional fields take nilable values but then set the default value (which does not get encoded), this was also surfaced by sorbet

[tenderlove]

"required" and "optional" refer to the encoding. If a field is "required", it means that it will always be encoded. If it's "optional", then it's only encoded if it is not the default value. (This is confusing from a user's perspective, but unfortunately that's how the spec is defined)

I think only fields that contain submessages are actually nillable. Everything else should require some kind of value.

[TobiasBales]

"required" and "optional" refer to the encoding. If a field is "required", it means that it will always be encoded. If it's "optional", then it's only encoded if it is not the default value. (This is confusing from a user's perspective, but unfortunately that's how the spec is defined)

I think only fields that contain submessages are actually nillable. Everything else should require some kind of value.

Interesting.
So should the constructor not accept nil values but rather have a default value then?
And setters (for optional fields) would then also just accept the non nilable type?
Doing that would simplify some things indeed