Restore ABI stability for annotatable and documentable builders

[ZacSweers]

Resolves #1579

Originally tried to be clever by using value classes to get the impl without the overhead, and it almost worked except that the resulting bytecode still tried to cast the underlying apply to the value class instead of the enclosing builder it was inlined into. Fell back to a more manual approach of just adding explicit overrides that just called super.

[Egorand]

Hmm, really confused why removing these redundant overrides breaks binary compatibility..

[JakeWharton]

Because the generic return type T winds up in bytecode as its lower bound which is the interface Builder rather than the concrete spec Builder.

Also the binary compatibility validator seems to be a complete garbage tool and doesn't fail the build when you do this. Do we need to switch to metalava or japicmp, or does it just have terrible default behavior?

[ZacSweers]

I think it focuses only on being a change detector and relies on code review to decide if the change is ok. It doesn't do what japicmp does as far as "is this compatible with a previous release" kind of checks, but it's also more accurate in some ways. Maybe there's a mode like that that I'm missing though

[ZacSweers]

Filed a FR here: Kotlin/binary-compatibility-validator#140

[Egorand]

Should we set up japicmp in the meantime?

[ZacSweers]

Seems like a good idea