fldread.F90: bad array allocation in the interpolation on the fly (in case land/sea mask in active)

[clem]

In the subroutine fld_interp (from fldread.F90), the size of a temporary array is retrieved before the array is even allocated:

         itmpi=SIZE(ztmp_fly_dta(jpi1_lsm:jpi2_lsm,jpj1_lsm:jpj2_lsm,:),1)
         itmpj=SIZE(ztmp_fly_dta(jpi1_lsm:jpi2_lsm,jpj1_lsm:jpj2_lsm,:),2)
         itmpz=kk
         ALLOCATE(ztmp_fly_dta(itmpi,itmpj,itmpz))

I would replace it by:

         itmpi=jpi2_lsm-jpi1_lsm+1
         itmpj=jpj2_lsm-jpj1_lsm+1
         itmpz=kk
         ALLOCATE(ztmp_fly_dta(itmpi,itmpj,itmpz))

Moreover, variable zfieldo is defined as allocatable but is not allocated and not used. It is therefore not needed

I would commit the necessary changes unless someone disagrees.