Extra lib directory when adding dependencies for PyTorchModel

[setu4993]

SDK Version: 2.3.0
PyTorch Framework version: 1.5
Python version: 3

When trying to deploy create a PyTorchModel object with dependencies, there's an extra intermediate lib directory being created.

Expected:

>>> opt/ml/code
>>>     |------ train.py
>>>     |------ common
>>>     |------ virtual-env

What's happening:

>>> opt/ml/code
>>>     |------ train.py
>>>     |------lib/
>>>                |------ common
>>>                |------ virtual-env

I found this line that causes this. Based on this, I think it should only occur for multi-model servers, but in my case it is happening while trying to use a single model.

[setu4993]

@metrizable : I think this should be tagged as a bug since documentation points to the expected being what users should expect.