support concurrent analytics / barrier

Consider a pipeline of independent analytics, each enriching some tuple with additional information:

```
readings<T> -> A1 -> A2 -> A3 -> results<R>
```

If the analytics are independent and non-trivial / time consuming, there's a desire to run the analytics concurrently for each tuple with a graph logically like:

```
/-> A1 ->\
readings<T> > |> A2 ->| -> result<R>
-> A3 ->/
```

I'm using the term "concurrent" instead of "parallel". Java8 Streams', and IBM Streams Java API's and SPL's, use of "parallel" is for "parallel-tuple processing": multiple tuples, each being processed in parallel in its own "channel" of the same replicated pipeline of operations.

I'm going to use a pull-request to capture API proposals/discussion.