Get CDF working as a configurable analysis.
This eliminates all but token dependence on VTKm in favor of VTK's vtkSortDataArray class (which uses vtkSMPTools). However, it has the advantage of handling different types of data arrays, whereas the previous VTK-m worklet insisted on being provided doubles. It still saves the CDF out as an array of floats regardless of the input array type, though.