*+Revise horizontal avg in Calc_kappa_shear_vertex

  Revised the Calc_kappa_shear_vertex code where it averages the diffusivities
at the vertices back to the tracer points when there is either a geometric mean
being used or thickness weighting.  With these changes, the code should now pass
dimensional consistency testing, work properly with openMP threading enabled,
and give sensible values of the diffusivity for massless layers.  This can
change answers slightly in some cases where VERTEX_SHEAR = True, especially if
VERTEX_SHEAR_GEOMETRIC_MEAN = True or VERTEX_SHEAR_THICKNESS_MEAN = True.  The
recently added runtime parameter VERTEX_SHEAR_KD_MEAN_ANSWER_DATE (which has not
yet been added to the dev/gfdl branch of the code) was no longer needed and has
been eliminated.

  Also revised the Calculate_kappa_shear diagnostics to work properly when
openMP threading is enabled, although this part of the changes does not change
any solutions.