A new, faster method for vtkPolyData::ComputeBounds()

Added new threaded methods in vtkPolyData and vtkBoundingBox
for computing GetBounds(). Added a comparison test for the
methods vtkPolyData::GetBounds(), vtkPoints::GetBounds(),
and vtkBoundingBox::ComputeBounds(). The speed improvements on
large meshes is approximately 10x.

Also, the new vtkCellArray API, and the vtkArrayDispatch
framework is used. Note, as before, this new vtkPolyData::GetBounds()
only considers points used by cells to determine the bounding box.

Performance considerations required tuning the vtkCellArray and
vtkCellArrayIterator to share memory when possible (i.e., avoid
copying). Also cleaned up documentation and API for these classes.
A test was added to ensure that iterator is functioning properly
(e.g., is faster, sharing storage buffer as appropriate).

Finally discovered an error in vtkParametricKuen which was
generating NaN on some systems, which messed up the GetBounds()
computation.