IBM: s_ibm_correct_state dereferences absent optional dummies (pb_in/mv_in) with no present() guard

[sbryngelson]

Bug

s_ibm_correct_state (src/simulation/m_ibm.fpp:143) declares pb_in/mv_in as optional:

real(stp), dimension(...), optional, intent(inout) :: pb_in, mv_in

but m_ibm.fpp contains zero present() calls. The routine uses them unconditionally. One of
its two call sites omits them (m_time_steppers.fpp:556):

if (qbmm .and. .not. polytropic) then
    call s_ibm_correct_state(q_cons_ts(1)%vf, q_prim_vf, pb_ts(1)%sf, mv_ts(1)%sf)
else
    call s_ibm_correct_state(q_cons_ts(1)%vf, q_prim_vf)      ! absent optionals
end if

That path is undefined behaviour. It happens to survive on some compilers; on Cray CCE 21 the
absent dummy presents a null dope vector to the device data environment and the run segfaults
(exit 139) on every IBM test.

Fix

pb_ts(1)%sf is always allocated — m_time_steppers.fpp allocates it in both branches
(full size when QBMM, beg:beg+1 otherwise), both with ACC_SETUP_SFs, and deallocates it
unconditionally. So it can be passed unconditionally:

1. drop optional from pb_in/mv_in in m_ibm.fpp:143
2. replace the if/else at m_time_steppers.fpp:553 with the single call

This is a net simplification (−4 lines) and removes the branch rather than making the absent
case survivable. Safe because the writes at m_ibm.fpp:437-438 are already guarded by
if (qbmm) / if (.not. polytropic), so the small non-QBMM allocation is never indexed.

Verification

Frontier (MI250X), CCE 21.0.2 / ROCm 7.2.0, full 627-test suite: zero exit-139 segfaults
after this change, where the IBM tests previously aborted. Found independently by two sessions.

Related

There is a second routine with the same shape that is not covered by this fix:
s_interpolate_image_point (m_ibm.fpp:807) has nine optional array dummies, and its call
sites pass different subsets — e.g. m_ibm.fpp:251 passes one and omits nine. That is a
suspect for a remaining IBM fault and should be audited the same way.