reuse virtual dataset instead of embeding results

For now I enforce the duplication of the dataset. This is due to a current limitation of h5pt virtual dataset that are only creating 'SimpleSelection' on VirtualSource.

The limitation is that we can have a virtual dataset (A) that we want to extend from an other virtual dataset(B) but only from a 'region' / slice of it. And this cannot be achieved yet.

In fact we should be able to retrieve the selection of the VirtualSource. But we are only getting those 'virtual source info' which are limiting. For now 'SimpleSelection' are created but we would need to use something like 'FancySelection' for now.