DocVec.from_protobuf(tensor_type=...) already exists, but this needs to be the case for all deserializations:

• proto
• json
• pandas
• bytes
• binary
• base64

Otherwise there is no way of knowing if the deserialized DocVec should use torch, np, or tf