Fixes a bug in evaluate cubic

In current CVMix master, the returned value of fprime in evaluate cubic
is not correct.  For example, when linear interpolation is used, the
returned value of fprime is

fprime = coeffs(2) + h + h^2

it should be

fprime = coeffs(2) + coeffs(3)*h + coeffs(4)*h^2

As it currently sits, the returned fprime is much too large (6 orders of
magnitude in my tests relative to the corrected version)

src/shared/cvmix_math.F90

@@ -244,7 +244,7 @@ function cvmix_math_evaluate_cubic(coeffs, x_in, fprime)
       cvmix_math_evaluate_cubic = cvmix_math_evaluate_cubic +                 &
                                   coeffs(i)*(x_in**(i-1))
       if (present(fprime).and.(i.gt.2)) &
-        fprime = fprime + real(i-1,cvmix_r8)*(x_in**(i-2))
+        fprime = fprime + coeffs(i)*real(i-1,cvmix_r8)*(x_in**(i-2))
     end do
 
   end function cvmix_math_evaluate_cubic