ENH: limit vtk floatField range (fixes #1055)

- for space-savings the VTK fields are normally written as 'float'
  rather than double. When a double field contains very large values,
  these can result in a overflow when converted to float.

  Now trap these with the appropriate numeric limits.
  No warning when these values are clipped: it should be readily
  apparent from the output.

ENH: handle symmTensor component swapping directly on VTK output.

- use VTK output routines in vtkSurfaceWriter to benefit from the
  above changes