fix tensor serialization

[numb3r3]

Maybe this problem due to the serialization of DA?

from jina import DocumentArray, Document, Flow
import numpy as np

docs = DocumentArray([Document(tensor=np.random.rand(224, 224))] * 100)
print(docs.summary())

with Flow(protocol='http').add() as flow:
    flow.post('/index', inputs=docs, show_progress=True)

with Flow(protocol='grpc').add() as flow:
    flow.post('/index', inputs=docs, show_progress=True)

yields

          Flow@96082[I]:🎉 Flow is ready to use!
	🔗 Protocol: 		HTTP
	🏠 Local access:	0.0.0.0:64451
	🔒 Private network:	192.168.31.35:64451
	💬 Swagger UI:		http://localhost:64451/docs
	📚 Redoc:		http://localhost:64451/redoc
⠼       DONE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╸ 0:00:48 100% ETA: 0 seconds 80 steps done in 48 seconds
           Flow@96082[I]:🎉 Flow is ready to use!
	🔗 Protocol: 		GRPC
	🏠 Local access:	0.0.0.0:49305
	🔒 Private network:	192.168.31.35:49305
⠋       DONE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╸ 0:00:00 100% ETA: 0 seconds 80 steps done in 0 seconds

We can observe that HTTP flow takes much more time.

[JoanFM]

I think there is very little to do about it.

[hanxiao]

closing as it is the wrong expectation of docarray. If client don't want embedding, they should not query embedding, eg. GraphQL