SPIKE - Investigate Apache Beam / Google Cloud Dataflow as a runner for custom ingest code

[danxmoran]

When working on #2 and #4 I set up a dumb CLP wrapper with hacky caching, which I've been running through the sbt shell. This is fine for prototyping, but not good for the long run.

Apache Beam has been repeatedly mentioned as a data-processing framework which might be well-suited to the type of tasks run during data ingest.

• Google provides Dataflow as a fully-managed runner for Beam, but it's not the only available solution.
• Spotify maintains a Scala wrapper (scio) for the Java Beam APIs.

Porting the ENCODE metadata-processing code to use scio and seeing how it runs through Dataflow should give us a good idea for what it's like to use Beam.