Modified to use dataset and pipelinedata

[GraemeMalcolm]

I modified the pipeline to be more "AMLish":

• Added instructions to Getting Started to register a dataset (or set variables for the pipeline to do it for you)
• Used input in training step to pass dataset object
• Added pipelinedata object to pass trained model as output from train step and input to register step

[GraemeMalcolm]

I'm getting a "Bad Gateway" error in the smoke test after the model service is deployed. I don't think this is due to my changes - it looks like an AML service error, but we should probably hold off on merging until we troublehoot it.

[dtzar]

Also, since we're no longer using the other two DATA env variables, you should remove them everywhere.