Small Bug in AGRIF (agrif_opa_interp.F90)

[gm]

In agrif_opa_interp, I think the subroutine interpun and interpvn has 2 small bugs (small as interpun, & vn are probably always call with k2=jpk).
This concerns both the trunk and the v3.6 stable
Two external do loops (jk-loop) have a wrong bound for example in interpvn:

   SUBROUTINE interpvn(ptab,i1,i2,j1,j2,k1,k2, before)
      !!---------------------------------------------
      !!   *** ROUTINE interpvn ***
      !!---------------------------------------------    
      !
      INTEGER, INTENT(in) :: i1,i2,j1,j2,k1,k2
      REAL(wp), DIMENSION(i1:i2,j1:j2,k1:k2), INTENT(inout) :: ptab
      LOGICAL, INTENT(in) :: before
      !
      INTEGER :: ji,jj,jk
      REAL(wp) :: zrhox	
      !!---------------------------------------------    
      !      
      IF (before) THEN          
         !interpv entre 1 et k2 et interpv2d en jpkp1
         DO jk=k1,jpk
            DO jj=j1,j2
               DO ji=i1,i2
                  ptab(ji,jj,jk) = e1v(ji,jj) * vn(ji,jj,jk)
                  ptab(ji,jj,jk) = ptab(ji,jj,jk) * fse3v(ji,jj,jk)
               END DO
            END DO
         END DO
      ELSE          
         zrhox= Agrif_Rhox()
         DO jk=1,jpkm1
            DO jj=j1,j2
               va(i1:i2,jj,jk) = (ptab(i1:i2,jj,jk)/(zrhox*e1v(i1:i2,jj)))
               va(i1:i2,jj,jk) = va(i1:i2,jj,jk) / fse3v(i1:i2,jj,jk)
            END DO
         END DO
      ENDIF
      !        
   END SUBROUTINE interpvn

DO jk=k1,k2 and DO jk=1,k2-1 should be used ?

[jchanut]

k1=1 in fact since vertical refinement is not allowed, so there is no consequence. For consistency with the rest of the code, let's replace k1 by 1 and restrict vertical loops to jpkm1 at the same time.
Done at revision 8633