net/openmpi: Fix 16 byte atomic operations on amd64

Clang does not support 16 byte atomic operations without -mcx16 on amd64
Upstream issue: https://github.com/open-mpi/ompi/issues/13134

PR: 285341
MFH: 2025Q1
(cherry picked from commit 07722f6ed2aa26e76ae830cfc26341788f4b7195)