bugs in "subroutine zgr_sco"

[hliu]

There are some bugs here in all the former versions of nemo I can see (include v3.1). These bugs can generate incorrect value of gdep3w which is used in dynhpg. The modifications are listed below:
1) ==================================================
!! H. Liu, Jan 2009
!! esigw( 1 ) = esigw( 2 )
!! esigt(jpk) = esigt(jpkm1)

esigw( 1 ) = 2.0_wp * (gsigt(1) - gsigw(1))
esigt(jpk) = 2.0_wp * (gsigt(jpk) - gsigw(jpk))

2) ==================================================

DO jk = 1, jpk

zcoeft = ( FLOAT(jk) - 0.5 ) / FLOAT(jpkm1)
zcoefw = ( FLOAT(jk) - 1.0 ) / FLOAT(jpkm1)
gdept (:,:,jk) = (scosrf(:,:)+(hbatt(:,:)-hift(:,:))*gsigt(jk)+hift(:,:)*zcoeft)
gdepw (:,:,jk) = (scosrf(:,:)+(hbatt(:,:)-hift(:,:))*gsigw(jk)+hift(:,:)*zcoefw)

!! R. Furnel and H. Liu
! gdep3w(:,:,jk) = (scosrf(:,:)+(hbatt(:,:)-hift(:,:))*gsi3w(jk)+hift(:,:)*zcoefw)

gdep3w(:,:,jk) = (scosrf(:,:)+(hbatt(:,:)-hift(:,:))*gsi3w(jk)+hift(:,:)*zcoeft)

END DO

===============================================================

N.B. The modification of esigt(jpk) is optional and will not affect model results. There is an attachment of the modified domzgr.F90.