Should Support the TupleUnpack/ListUnpack in TorchModuleGraph

[zheng-ningxin]

Environment:

• NNI version: master(latest)
• NNI mode (local|remote|pai): local
• Server OS (for remote mode only): Ubuntu16.04
• Python version: 3.7
• PyTorch/TensorFlow version: 1.4.0
• Is conda/virtualenv/venv used?: yes
• Is running in Docker?: no

What issue meet, what's expected?:
Because NNI does not have special treatment for TupleUnpack and ListUnpack operations, TorchModuleGraph will have a graph construction error in some models with TupleUnpack and ListUnpack operations(such as retina face/shufflenet). I want to support these operations in the next release and fix this problem.

How to reproduce it?:
Take shufflenet as an example. The TorchModuleGraph will build the graph in the wrong way(as shown in the following img).

In contrast, the right way to build the graph of the shufflenet is shown in the following graph.

Additional information:

[zheng-ningxin]

#2609