Support for audience segments in AB tests

While it's possible to see if a single title or headline for example work better across the board it will provide useful insights to be able to get reports on specific segments. A test variant that works better for one group may not work as well for another.

For example desktop viewers may convert better with longer headlines vs mobile viewers.

Additionally an A/B test may only be desired to run on mobile devices or for some other specific audience segment.

Two things to look into here:

- Reporting by segment
- Testing by segment

Related analytics issue: humanmade/analytics#19