Provide a way for a publisher to set up experiments in which content on the page is changed according to an experiment group a reader is placed in.

This is forked from an idea proposed in #945 (comment):

In the future a-b testing can be implemented [using dynamic CSS classes] as well, where we define a way to express rules for how to add such classes and publishers can say e.g. "add test-group-a for 10% of users" or a similar expression.

cc @cramforce @jridgewell