bug: m_cbc.fpp AMD local arrays hardcoded to dimension(3), overflow when num_fluids > 3

[sbryngelson]

Summary

In src/simulation/m_pressure_relaxation.fpp, the non-case-optimization AMD branch (#:if not MFC_CASE_OPTIMIZATION and USING_AMD) declares local arrays with a hardcoded upper bound of 2:

#:if not MFC_CASE_OPTIMIZATION and USING_AMD
    real(wp), dimension(2) :: alpha_rho, alpha
#:else
    real(wp), dimension(num_fluids) :: alpha_rho, alpha
#:endif

These arrays are then written inside a loop up to num_fluids:

do i = 1, num_fluids
    alpha_rho(i) = q_cons_vf(i)%sf(j, k, l)
    alpha(i) = q_cons_vf(eqn_idx%E + i)%sf(j, k, l)
end do

Impact

When num_fluids > 2 in an AMD GPU build without case optimization, both arrays overflow. The existing PROHIBIT checks do not specifically guard num_fluids <= 2 on AMD paths.

(Note: m_cbc.fpp has a similar AMD-specific hardcoded-dimension pattern but uses dimension(3) there.)

Fix

Replace the hardcoded dimension(2) with dimension(max(2, num_fluids)) or add a @:PROHIBIT(num_fluids > 2 ...) guard, or use the same dimension(num_fluids) path as non-AMD.

Discovered

Found during review of PR #1365. Pre-existing in master.