Undefined values in FCT tracer advection scheme when using MLF and "accurate" adaptive-implicit vertical advection algorithm

Context

• Branches impacted: branch_5.0
• Dependencies: ln_traadv_fct=.true., ln_zad_aimp=.true., nn_fct_imp=2, MLF

Analysis

In tridia_solver, all loops have bounds that include the first halo points. This means the input array arguments must also be defined on the first halo points.

In fct_up1_1stp when ll_zAimp2=.true., the input array arguments (zwdia, zwsup, zwinf) are defined only on internal points, so tridia_solver accesses undefined points in this case.

DO_3D( 0, 0, 0, 0, 1, jpkm1 )
   zwdia(ji,jj,jk) = 1._wp + pDt * ( MAX( wi(ji,jj,jk  ) , 0._wp ) - MIN( wi(ji,jj,jk+1) , 0._wp ) ) / e3t(ji,jj,jk,Kaa)
   zwinf(ji,jj,jk) =         pDt *   MIN( wi(ji,jj,jk  ) , 0._wp )                                   / e3t(ji,jj,jk,Kaa)
   zwsup(ji,jj,jk) =       - pDt *   MAX( wi(ji,jj,jk+1) , 0._wp )                                   / e3t(ji,jj,jk,Kaa)
END_3D
!
CALL tridia_solver( zwdia, zwsup, zwinf, pt_up1, pt_up1 , 0 )

This is also true of ztmp on line 288.

fct_up1_1stp is called only by MLF, so this does not affect RK3. There is therefore no issue in the main.

Fix

Add an option to specify the bounds for loops in tridia_solver and set these accordingly.

This should be done for the main branch as well as branch_5.0 even though there is no issue in the main, because the proposed changes will reduce unnecessary calculations.

Other notes

fct_up1_1stp is no longer used- it should be removed from the main at some point.