Minor fixes for distributed contour tree example

Various minor fixes to distributed contour tree example (mainly to get it running with CUDA support on Perlmutter @ NERSC)

• Add missing MODIFY_CUDA_FLAGS DEVICE_SOURCES directives to target information for TreeCompiler and BranchCompiler
• Add a flag to enable use of a parallel HDF5 library with vtk-m, as the distributed contour tree example needs a parallel HDF5 implementation and there were compiler problems when compiling on Perlmutter @ NERSC with HDF5 and CUDA support. The new option VTKm_HDF5_IS_PARALLEL ensures that sources dependent on HDF5 will also be compiled with MPI includes in the search path.
• Clean-up by removing unnecessary WITH_MPI define and changing default value type to use double precision floating point.