Introduced new composite datasets for partitioned data.

This is the first step in our attempt to create next
generation composite datasets for partitioned data and
arbitrary dataset collections. The main idea is that
partitioned datasets are instances of
vtkPartitionedDataSet, which is a vector of datasets.
It is required that these datasets are compatible, i.e.
can be processed by the same type of filters together.
This is similar to vtkMultiPieceDataSet but will
formalize that partition concept further. The second
new data type is vtkPartitionedDataSetCollection. This
is a vector of vtkPartitionedDataSets. It can be an
arbitrary collection but is not a tree like its
predecessor vtkMultiBlockDataSet. Any further grouping
of the blocks will be done by the use of SIL data
structures, which will be supported later.

Also added basic IO support for these data types.