Bug from removing cvmix_kpp_compute_nonlocal

I looked for a parameter in CVmix_kpp_params_user when I should have been
looking in CVmix_kpp_params_in... was causing a seg fault in some POP testing.
I don't know why the regression test suite didn't catch this...

Author: mnlevy1981

src/shared/cvmix_kpp.F90

@@ -904,9 +904,9 @@ subroutine cvmix_coeffs_kpp_low(Mdiff_out, Tdiff_out, Sdiff_out, zw, zt,    &
       !   (3c) Compute nonlocal term at each cell interface
       if ((.not.lstable).and.(kw.le.kwup)) then
         GAtS = cvmix_math_evaluate_cubic(Tshape2, sigma(kw))
-        Tnonlocal(kw) = CVmix_kpp_params_user%nonlocal_coeff*GAtS
+        Tnonlocal(kw) = CVmix_kpp_params_in%nonlocal_coeff*GAtS
         GAtS = cvmix_math_evaluate_cubic(Sshape2, sigma(kw))
-        Snonlocal(kw) = CVmix_kpp_params_user%nonlocal_coeff*GAtS
+        Snonlocal(kw) = CVmix_kpp_params_in%nonlocal_coeff*GAtS
       end if
 
       !   (3d) Diffusivity = OBL_depth * (turbulent scale) * G(sigma)