Make the SDK more user-friendly by specifying configuration defaults

[marcelgwerder]

We're trying to make the SDK easier to use for colleagues that don't really want to deal with cloud-specific configuration such as IAM roles, KMS keys etc. but instead just want to focus on running their experiments as jobs or deploy their model as an endpoint. Additionally, we have certain naming and tagging requirements that we would like to cover as well.

How we've approached this so far is that we would provide some factory classes that are aware of the configuration details and then would create the SDK objects:

factory = SageMakerFactory.load() # load from environment as an example
# factory.workspace.iam_role
# factory.workspace.naming_prefix 
# factory.workspace.tags
# etc.

factory.create_estimator(base_job_name="my-job", instance_type="...")

And the factory would prefix the job name and fill in the gaps such as tags, role, kms key, etc.

This generally works fairly well but doesn't cover cases such as Pipeline.upsert where the information is not provided upon object creation but rather when calling some method on it. Ideally, we want to avoid extending the SDK too much as this would become a maintenance nightmare. Is this kind of default configuration something you would consider adding support for, or at least allow providing these defaults upon object creation? Currently it is fairly inconsistent, the Estimator would, for example, take the tags in the constructor, yet Pipeline only takes the argument when calling upsert.