Hotfix: Fix RHEL patch for C11 non-RHEL builds

[amd-vserbu]

Summary

scripts/package-ami.sh rewrites the eventfd_signal() version gate in submodules/AVED/sw/AMI/driver/ami_program.c so the void-arg form is also selected on RHEL 9.5+, where Red Hat backported the upstream 6.8 simplification into the 5.14-based kernel.

The current rewrite uses a single composite #if:

#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0) \
 || (defined(RHEL_RELEASE_CODE) \
     && RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(9, 5))

This works on RHEL but breaks the build on other distros with kernel >= 5.18 (Ubuntu 24.04, Ubuntu 22.04 HWE), if the kernel is compiled under C11 (-std=gnu11, the default since v5.18).

Why it breaks

Per C11, in a #if controlling expression all identifiers remaining after macro expansion (other than the operand of defined) are replaced with the pp-number 0, including function-like macro names. On a non-RHEL kernel RHEL_RELEASE_VERSION is undefined, so the preprocessor ends up parsing:

... || (0 && 0 >= 0(9, 5))
                   ^^^^^^^ syntax error

i.e. MACRO(a, b) becomes 0(a, b), which is not a valid constant expression. The defined(RHEL_RELEASE_CODE) short-circuit doesn't help, the rule applies during expression parsing, before evaluation.

Fix

Replace the sed rewrite with a patch that guards the RHEL check behind #ifdef RHEL_RELEASE_CODE, so non-RHEL translation units never tokenize RHEL_RELEASE_VERSION in a #if at all:

#if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 8, 0)
    eventfd_signal(efd_ctx);
#else
# ifdef RHEL_RELEASE_CODE
#  if RHEL_RELEASE_CODE >= RHEL_RELEASE_VERSION(9, 5)
    eventfd_signal(efd_ctx);
#  else
    eventfd_signal(efd_ctx, <count>);
#  endif
# else
    eventfd_signal(efd_ctx, <count>);
# endif
#endif

Applied via patch --no-backup-if-mismatch -p1 -d "${AVED_DIR}"; the existing trap already git checkouts ami_program.c on exit, so revert semantics are unchanged.