Add workspace semantics to tf-estimators notebook

[zomglings]

Handling directories through a Jupyter notebook is kind of awkward.

@amygdala had a good suggestion:

one thought about the model_dir -- in some of my notebooks I've been doing something like this:
model_dir = os.path.join(base, name + "_" + str(int(time.time())))
where 'base' in your case I guess is the setup dir. That way, they can do multiple runs of the notebook, then point TensorBoard at the parent 'base' dir and compare runs across the timestamped subdirs.
I don't know if that would make sense here, and might not work with the other ways you're using the setup dir.

This indicates a general workflow:

1. User specifies a base directory/workspace in which our code can create as much stuff as it wants.

2. When the user is done with the contents of the notebook, they know exactly what to get rid of from their filesystem.

Action item

Make the tf-estimators notebook consistent with this workflow.

[zomglings]

Not worth the effort.