The previous busy-loop fix did not stop the flake: the test still timed
out after 30s waiting for the peer handshake. Root cause is the endpoint
address. The peers pointed at each other via the host's routable NIC IP
(getLocalIP), so the handshake had to hairpin through the host/container
network stack. Under CI load that drops packets, and WireGuard only
retries a lost handshake initiation every REKEY_TIMEOUT (5s), stacking up
to the ~30s timeout.
Point both endpoints at 127.0.0.1 instead. The userspace WireGuard bind
listens on 0.0.0.0, so loopback reaches it over a path that never drops.
Remove the now-unused getLocalIP helper.
34d640dfc9