CI outerloop queues are forced to run sequentially

Our CI outerloops rely on us calling an HTTP endpoint to sync server code, after which the client will test against the assigned server.

At the moment, we have a design limitation where one server endpoint is assigned to one CI queue; if we allow concurrent builds, it becomes possible to pave over the server endpoint with mismatched code while another test is running.

Per https://github.com/dotnet/wcf/pull/1064#discussion_r60630840, @mmitche says:

> It seems like we need a better solution for this eventually. I can see lots of places where we would want to deploy multiple servers (developer scenarios for instance) concurrently.

Investigate how we might do this