Convert to new split pipeline approach for CI/CD

[j-so]

Introduces a new set of pipelines that split model/train/eval and deployment.

1. Model training/eval/registration - performs Model CI, trains, evaluates, and registers the model. If the model is registered, then it produces a pipeline artifact containing the model JSON.
2. Model deployment - takes the pipeline artifact (model JSON) from the triggering pipeline and deploys that model.

NOTE: There is a limitation with the model deployment pipeline. It relies solely on the output of the model training pipeline. Right now it cannot run off the model name/version env vars.

Also addresses #274

[dtzar]

Since the CI steps are identical, I believe we should keep thing consistent by having a CI and then a CD pipeline which is triggered when the CI pipeline is complete. Sooo... This means for this PR:

1. No need for a separate "getting started doc"
2. Remove duplicate CI pipeline, fix the existing one to move the CD steps into a "Task based deployment" pipeline.
3. Fix the main getting started guide to ask people if they want the AzDO task deployment or Az CLI deployment, setting whatever you feel is the best default one triggered by the CI pipeline. If you feel the engineering team doesn't want people to have the task deployment, then go ahead and remove this choice.