Add back CustomDebugStringConvertible conformance

[gjcairo]

This is part of a wider effort to make breaking changes introduced in main non-breaking.

#1246 removed the inheritance of CustomDebugStringConvertible from AnyExtensionField, and instead moved conformance on to the concrete types (OptionalExtensionField, RepeatedExtensionField, etc), but wrapping these extensions in #if DEBUG statements to avoid unnecessary large binary sizes in release mode.

This was a breaking change, so we need to find a compromise to make this non-breaking but get some of the release mode binary size improvements.

This PR moves the implementation of the conformance of several types to the CustomDebugStringConvertible protocol inside #if DEBUG clauses, instead of the whole conformance declaration. We won't get the same size savings since the conformances will still be captured by the runtime, but we won't get any breaking changes.

[thomasvl]

Any chance you can get numbers for a release build before/after this PR on something with some protos so we can gauge the change in size?

[gjcairo]

I created an otherwise empty executable that contains kubernetes' API protobufs (https://github.com/kubernetes/api/blob/master/core/v1/generated.proto) which is quite sizeable. I built in release mode on Linux, using Swift 6.

• If I revert the change introduced in Conform to CustomDebugStringConvertible for Debug only #1246 altogether, the binary size is 20.7MB.
• With the changes on main, the binary size is 19.3MB (that's ~8.75% decrease).
• With the changes on this PR, the binary size is 20MB (that's a ~3.4% decrease).
• I also tried annotating the debugDescription implementations with @inlinable and @_alwaysEmitIntoClient, but that yielded binaries of around 20.1MB (~2.9% decrease).

Most of this comes from the fact that the Message protocol inherits from CustomDebugStringConvertible, and generated code conforms to Message.
This means that if we merge this to avoid the API breakage, we'd be losing just over half of the size decrease benefit of the original change (from 8.75% to 3.4%).

I know this isn't ideal, but we are still reducing the binary size compared to the latest release. We can revisit this if/when we release a v2 eventually.

[FranzBusch]

LGTM! This is a fair compromise for now to avoid the API breakage. If we tackle a v2 in the future we should check if we can remove all these public protocols like Enum and/or remove all the inherited protocols from it and just do this on the code generation side. This will allow us to give the generator options to include various conformances and end-users can fine tune these options to their needs.