Make use of HDF5 virtual datasets

[abensonca]

When output is written to multiple files (e.g. under MPI), it might be helpful to generate a virtual datasets in one of the files (e.g. the MPI0000 file) that link together datasets in all other files. These could be in some subgroup, e.g.:

Outputs/Output1_all/nodeData

and Outputs/Output1_all should probably have a copy of the mergerTreeStart etc dataset that has the corrected offset information into these virtual datasets. (The other datasets, e.g. mergerTreeIndex can just be virtual datasets).

This would remove the need for (significantly) re-writing code when using MPI as the virtual datasets can just be read.