Simplify input to AutoMLStep

[dataders]

This is something I did a few weeks ago as I was debugging my own pipeline. For me, this is a a clearer and more robust way to do train test split w/ AutoMLStep and PipelineData objects. The changes I made:

1. using os.path.join() instead of + to concat strings
2. testTrainSplitStep has 1 `PipelineData object instead of 4
   @sanpil @yanrez @j-martens let me know what y'all think

[msftclas]

All CLA requirements met.

[purnesh42H]

Don't include config.json file

[dataders]

I made some commits to make it fit the existing repo. As I understand it, this isn't the real repo though so does getting this approved even matter? I'm just proposing changes to this notebook. @rastala

[purnesh42H]

A check is needed to see if output is specified(not None) as the steps don't enforce it.

[sanpil]

@swanderz Please continue to propose changes. Even though the publishing mechanism is different, we could adopt your changes.

[dataders]

@sanpil this PR is my feedback, can you please give feedback on the version of the notebook I'm proposing (not sure how to get see a helpful diff on .ipynb's in this PR)

Let me know what you think.

If you understand and agree with my changes, I'm then willing to work with @purnesh42H and @rastala to get the PR approved, but we're definitely not at that point yet.

[sanpil]

@purnesh42H and I will review the changes and will get back to you. Thanks @swanderz!