fix(exp_006): handle systemd port conflict — add Preexisting AMF runtime variant

[Copilot]

When open5gs-amf installs via apt, systemd auto-starts open5gs-amfd, pre-binding SBI port 7777 and Prometheus port 9090. The experiment then spawned a second instance that exited immediately, leaving AmfRuntime::NativePid holding a dead Child while silently falling through to the systemd-managed instance — no diagnostic, no clean ownership.

Changes

experiments/exp_006_open5g_core_comparison/run.rs

• New AmfRuntime::Preexisting variant — models "instance already running, we don't own it"; Drop is a no-op
• Immediate-exit detection — after spawn, polls child.try_wait() every 100 ms for up to 1 s; if the process exits, logs the exit status and falls back to Preexisting rather than holding a zombie Child
• Metrics address detection (detect_metrics_addr) extracted before the match — eliminates the duplicated call across branches

let immediate_exit = (0..10).find_map(|_| {
    std::thread::sleep(Duration::from_millis(100));
    child.try_wait().ok().and_then(|s| s)
});
let runtime = match immediate_exit {
    Some(status) => {
        println!("Binary exited immediately (status={status}); using pre-existing open5gs AMF instance");
        AmfRuntime::Preexisting
    }
    None => AmfRuntime::NativePid(child),
};

.github/workflows/open5gs-integration.yml

• New step: "Stop systemd open5gs-amfd service" — runs systemctl stop/disable open5gs-amfd + 1 s port-release pause before the experiment, so CI always exercises the NativePid path with a fresh controlled instance
• Preexisting remains as a graceful fallback for local dev where systemd cannot be stopped

Original prompt

---

This section details on the original issue you should resolve

<issue_title>test with actual open5gc vs our 6g system</issue_title>
<issue_description>```
=== exp_006: open5GS Actual System Integration Test ===
[Step 0] Native open5gs binary detected: /usr/bin/open5gs-amfd
Config: /etc/open5gs/amf.yaml
Open5GS daemon v2.7.6
02/28 15:55:59.304: [app] INFO: Configuration: '/etc/open5gs/amf.yaml' (../lib/app/ogs-init.c:144)
02/28 15:55:59.304: [app] INFO: File Logging: '/var/log/open5gs/amf.log' (../lib/app/ogs-init.c:147)
02/28 15:55:59.307: [sbi] INFO: Setup NF EndPoint(addr) [127.0.0.200:7777] (../lib/sbi/context.c:507)
Failed to bind to port 0: Address already in use
02/28 15:55:59.307: [metrics] ERROR: Cannot start Prometheus HTTP server (../lib/metrics/prometheus/context.c:283)
02/28 15:55:59.307: [sbi] INFO: NF Service [namf-comm] (../lib/sbi/context.c:1994)
02/28 15:55:59.307: [sock] ERROR: socket bind(2) [127.0.0.5]:7777 failed (98:Address already in use) (../lib/core/ogs-socket.c:114)
02/28 15:55:59.307: [sock] ERROR: tcp_server() [127.0.0.5]:7777 failed (98:Address already in use) (../lib/core/ogs-tcp.c:72)
02/28 15:55:59.307: [sbi] ERROR: Cannot start SBI server (../lib/sbi/nghttp2-server.c:414)
02/28 15:55:59.307: [sctp] ERROR: Failed to initialize AMF (../src/amf/app.c:30)
02/28 15:55:59.307: [app] FATAL: Open5GS initialization failed. Aborted (../src/main.c:224)
Metrics endpoint: 127.0.0.5:9090
[Step 1] Actual open5gs AMF configuration (open5gs (native))
PLMN: 999/70
TAC: 1
SST: 1
SBI port: 7777
[Step 2] Live open5gs AMF metrics (no UEs attached yet)
fivegs_amffunction_rm_reginitreq : 0
fivegs_amffunction_rm_reginitsucc: 0
fivegs_amffunction_amf_authreq : 0 ← per-UE auth cost (TS 33.501 §6.1)
amf_session: 0 gnb: 0
[Step 3] 6G SBAv2 simulation — PLMN 999/70 TAC 1 SST 1 (5 UEs)
UE=999700000000001 session_id=1 ip=10.0.0.1 qci=9 gbr=0kbps
UE=999700000000002 session_id=2 ip=10.0.0.2 qci=9 gbr=0kbps
UE=999700000000003 session_id=3 ip=10.0.0.3 qci=9 gbr=0kbps
UE=999700000000004 session_id=4 ip=10.0.0.4 qci=9 gbr=0kbps
UE=999700000000005 session_id=5 ip=10.0.0.5 qci=9 gbr=0kbps
[Step 4] NAS overhead comparison
open5gs live (after sim): reginitreq=0 authreq=0
Projected 5G NAS cost for 5 UEs (open5gs schema):
reginitreq = 5 (one per UE)
authreq = 5 (one per UE — TS 33.501 §6.1)
6G SBAv2 actual result for 5 UEs:
validated = 5 (SBAv2 inline token — 1 RTT)
authreq = 0 (eliminated — token inline in first PDU)
authreq_reduction = 100%
[Step 5] [baseline: open5gs (actual binary) / registration_success_rate] citation: https://github.com/open5gs/open5gs
All 1 comparison points within tolerance ✓
open5gs live: PLMN 999/70 TAC 1 SST 1 (open5gs (native))
6G SBAv2: 5 UEs, 5 sessions, 67 B UPF uplink
NAS overhead: 5G NAS = 1/UE auth RTT → 6G SBAv2 = 0 (100% reduction)
All exp_006 checks PASSED ✓

<agent_instructions>this experiment is facing the issue. the actual open5gs doesn't seem to work. and actual 6g implementation actual call flow comparison.</agent_instructions>

## Comments on the Issue (you are @copilot in this section)

<comments>
</comments>

• Fixes test with actual open5gc vs our 6g system #25

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.