independent handling of MPI_Request

The current handling of MPI_Request is generally based on the idea of handling multiple requests and generally has this type of coding:

const label startOfRequests = UPstream::nRequests();

... setup receives
... setup sends

UPstream::waitRequests(startOfRequests);

However, when we start wishing to intersperse individual non-blocking transfers, we end up with this type of stuff:

    if
    (
        outstandingRequest >= 0
     && outstandingRequest < UPstream::nRequests()
    )
    {
        UPstream::waitRequest(outstandingRequest);
    }

The integer value of outstandingRequest is from somewhere within the overall list of requests, and internally there is various bits of logic to recycle the freedRequests, with side-effect of allToAll communication being used to re-align things. In summary: a bit of a mess.