test(tailor): assure weights are preserved after calling to_embedding_model

[bwanglzu]

test after calling to_embedding_model, weights are preserved.

Note:

1. default behaviour keras and (paddle, torch) are different, weights in keras are randomly initialised. So in keras only tested pre-trained vgg16 model.
2. only tested first layer, as not so much difference in the later layers, testing the first layer of weights is sufficient enough to prove weights did not change