Fix environment factory pickling on Windows

[harperj]

SubprocessUnityEnvironment sends an environment factory function to
each worker which it can use to create a UnityEnvironment to interact
with. We use Python's standard multiprocessing library, which pickles
all data sent to the subprocess. The built-in pickle library doesn't
pickle function objects on Windows machines (tested with Python 3.6 on
Windows 10 Pro).

This PR adds cloudpickle as a dependency in order to serialize the
environment factory. Other implementations of subprocess environments
do the same:
https://github.com/openai/baselines/blob/master/baselines/common/vec_env/subproc_vec_env.py

[xiaomaogy]

LGTM

[xiaomaogy]

@harperj Just to make sure, does the model get saved under Windows for these two cases?

1. ctrl + c interrupt.
2. Finish the training normally.

Also have you tested the Linux environment for these two cases?

[SetoKaiba]

1. Stop from Unity Editor.
   Now: Model is not saved.
   Previous: Model is saved. It's saved even just after the training begins for seconds.

2. Run out the whole training for max_steps.
   Now: The Unity Editor is frozen. And the training is never finished. The checkpoints are saved. But the nn model is not. Press Ctrl C and the Unity Editor is not frozen any more. The nn model is saved.
   Previous: The Unity Editor will stop playing. The training is finished. Both checkpoints and nn model are saved. You don't have to press Ctrl C.

3. Stop with Ctrl C.
   Now: Model is not saved.
   Previous: Model is saved. It's saved even just after the training begins for seconds.

Test Environment: Windows 10

[xiaomaogy]

@SetoKaiba Thanks for testing this for us! @harperj Just showed me a fix yesterday afternoon, which works for case 2 and case 3 now, I believe he will test the case 1 and then update this branch.

[xiaomaogy]

Changes needed to fix the ctrl+c and max_step reached errors.

[harperj]

@xiaomaogy @SetoKaiba Those are different (unrelated) issues which I'll shortly make subsequent PRs for.