"join" steps internally in ES

[rmagen]

This approach could reduce traffic and extra hops. it is especially useful if you model your data so that the adjacent steps are routed to the same node. just make sure the routed shard (or a replica) are on the same node.

We could use something like this: elastic/elasticsearch#3278. This PR creates "only" one join at a time, but it could be developed to do more.

Alternatively, if we do not want to wait for that, we could create a plugin with the same functionality.