refactor(tailor): improve interface

[hanxiao]

No description provided.

[bwanglzu]

what has been done in this PR?

1. api change, user provide name instead of idx.
2. adjust unit tests
3. now pytorch embedding_layer_name has changed from Class-Number to class_number.

How embedding_layer_name looks different between pytorch and paddle?

1. paddle has built-in layer _full_name, while pytorch not. Paddle layer name starts from 0, in pytorch we start from 1
2. paddle [layer_name]_[idx], the idx increase 1 with respect to the same layer_name, e.g. linear_0, relu_0, linear_1, while pytroch looks like this: linear_1, relu_2, linear_3.