BP marshaling in SST requires extra global MPI operations
SST has a need to aggregate metadata from different ranks to provide it to the reader ranks. SST's metadata includes SST-based metadata in addition to the metadata associated with written application data. For FFS marshaling, we piggyback aggregating FFS-based metadata along with SST-based things using the same global operations. Unfortunately the BP3Serializer operation AggregateCollectiveMetadata() is does it's own MPI global operations, so we cannot use it in such a way that we can get by with the same global operations for both sorts of aggregation. Ideally AggregateCollectiveMetadata() would be augmented with different routines that would assemble aggregate metadata using individual rank metadata that SST has transported to rank 0 (or some other rank).