Parallel operator stats only relate to one child

Currently, the stats for the parallel execution operator are displayed for only one copy of the child, not all.
This is currently coded like this because in this way, the plan tree and the execution tree match.
While this gives a good idea of how much of a bottleneck the child operator is, and therefore is good to go as far as the feature is concerned, this is not precise as it could be.
We have several options
1- leave things as they are (imprecise)
2- break the matching between plan and execution, and have plan report 1 child, execution report all children (could affect the UI)
3- change both plan and execution to report all copies of the child (confusing as far as plan goes, might break test runner and UI)

The fourth option - which is to aggregate all children is expensive to execute (we have to create an extra copy of the child and aggregate the stats in there, marshal it and then release the extra child), has memory implications and only works if the child is a leaf, so I think this is a no go.