codec-protobuf: a protobuf JsonCodec sibling for byte[] systems (e.g. Kafka)

[dfa1]

JsonCodec now speaks byte[] (#35), so a binary codec fits next to the JSON ones as a codec-protobuf module.

To decide

• Schema: a .proto for the model. Content.Fields is an arbitrary JSON object (Map<String, Object>), so it needs google.protobuf.Struct/Value and a mapping both ways; Answer/Question map to oneof.
• SPI: JsonCodec is named for JSON, and client-http must keep requiring JSON (the API speaks JSON). Either a protobuf codec stands alone (a ProtobufCodec class used directly, e.g. by a Kafka serializer), or the SPI splits into a generic Codec with JsonCodec as the subtype client-http asks for. Only worth splitting if something needs to be codec-agnostic.
• Who consumes it: a concrete producer/consumer should drive the schema before we freeze it.

🤖 Generated with Claude Code

[dfa1]

Update: the SPI is now Codec (byte[] in/out) as of #35, so a protobuf codec implements it directly. Remaining open point: client-http must still pick a JSON codec, since the API speaks JSON. If a non-JSON codec can be on the classpath, Codec needs a way to say its format (e.g. a media type) so ServiceLoader discovery in client-http can filter.