Feature/operations actually

[drewbanin]

Fixes #582, at least initially. This is a feature that we need to deploy and play around with to understand better, but it's difficult to use it for production workloads if it's not in a production release of dbt!

I suspect this functionality may change in meaningful ways, and I therefore suggest that:

1. we discourage the use of the task for folks who aren't interested in bleeding-edge dbt features
2. we do not document or promote this feature until (at least) the next release of dbt

I tried hiding the subtask from the argparse help menu, but i think argparse.SUPPRESS only works for flags (and not subtasks).

Once this is merged and we've had some time to experiment with it, we should queue up a subsequent issue to round out any rough edges and add some tasks.

Potential use cases:

• vacuum/analyze on redshift
• auto-compression for big tables on redshift
• periodic unloads from redshift
• copy into snowflake from s3
• swap schemas (eg. for blue/green deployments)
• really any out-of-band SQL that should be templated + versioned is suitable here

Really compellingly, we can supply entrypoint macros in packages, so the redshift package could contain macros that make it dead simple to do some really complex perf tuning!

[cmcarthur]

looks ok to me. i left you one little comment.