So far we didn't really take care of serialization and replication for RedisAI data types.
It should be straightforward, with one caveat: in MODELSET we take the protobuf, deserialize it and throw it away (we only keep the model in memory, not the protobuf as it would be wasteful).
However, we will need to serialize the protobuf as-is (pre deserialization) for persistence and replication within the call to ModelSet.
Related to #63.