fix: resolve Frontier CCE 3D cylindrical test failures

Two root causes for 5 failing 3D Cylindrical tests on Frontier CCE:

1. Post-process segfault (non-viscous tests 301B9153, 128954AD):
   s_axis() declared pb_in/mv_in as non-optional, but they are passed
   as absent optionals from s_populate_variables_buffers in post_process.
   CCE dereferences the null descriptor; gfortran tolerates it. Added
   'optional' attribute and present() guards, matching peer routines
   like s_symmetry and s_periodic.

2. Simulation FPE (viscous tests 07C33719, 939D6718, 09623DE3):
   The direct MPI execution path in the test runner bypasses the batch
   template which sets 'ulimit -s unlimited'. CCE-compiled viscous
   cylindrical binaries overflow the default stack. Set the soft stack
   limit to the hard limit at module load via resource.setrlimit.

Author: Spencer Bryngelson

FRONTIER_CCE_POST_PROCESS.md

@@ -763,7 +763,7 @@ contains
     subroutine s_axis(q_prim_vf, pb_in, mv_in, k, l)
         $:GPU_ROUTINE(parallelism='[seq]')
         type(scalar_field), dimension(sys_size), intent(inout) :: q_prim_vf
-        real(stp), dimension(idwbuff(1)%beg:, idwbuff(2)%beg:, idwbuff(3)%beg:, 1:, 1:), intent(inout) :: pb_in, mv_in
+        real(stp), optional, dimension(idwbuff(1)%beg:, idwbuff(2)%beg:, idwbuff(3)%beg:, 1:, 1:), intent(inout) :: pb_in, mv_in
         integer, intent(in) :: k, l
 
         integer :: j, q, i
@@ -804,7 +804,7 @@ contains
             end if
         end do
 
-        if (qbmm .and. .not. polytropic) then
+        if (qbmm .and. .not. polytropic .and. present(pb_in) .and. present(mv_in)) then
             do i = 1, nb
                 do q = 1, nnode
                     do j = 1, buff_size

src/common/m_boundary_common.fpp

@@ -6,6 +6,7 @@
 import itertools
 import json
 import os
+import resource
 import shutil
 import subprocess
 from typing import Callable, Dict, List, Optional, Set, Union
@@ -15,6 +16,14 @@
 from ..run import input
 from ..state import ARG
 
+# MFC simulations (especially viscous cylindrical cases) require large stack
+# sizes.  The batch templates all include `ulimit -s unlimited`, but the direct
+# MPI execution path in the test runner bypasses the templates.  Set the soft
+# stack limit to the hard limit so that spawned MPI processes inherit it.
+_soft, _hard = resource.getrlimit(resource.RLIMIT_STACK)
+if _soft != resource.RLIM_INFINITY and _soft < _hard:
+    resource.setrlimit(resource.RLIMIT_STACK, (_hard, _hard))
+
 
 @dataclasses.dataclass
 class MPIConfig: