s2idle wakes immediately after suspend on GZ302/Strix Halo — regression 7.1.6→7.1.8

## Summary

s2idle wakes the system 2-6 seconds after suspend, unprompted, on ROG Flow Z13 (GZ302, Strix Halo). Regression between linux-cachyos 7.1.6-1 and 7.1.8-1. Not present on linux-cachyos-lts 6.18.42-1 on the same machine/BIOS.

## System

- Model: ASUS ROG Flow Z13 GZ302EA (Strix Halo)
- BIOS: GZ302EA.311 (2025-08-01), latest available from ASUS
- Kernel (broken): linux-cachyos 7.1.8-1
- Kernel (working): linux-cachyos-lts 6.18.42-1
- Last known-good: linux-cachyos 7.1.6-1 (upgraded to 7.1.8-1 on 2026-08-10, broken sleep observed starting the next boot, 2026-08-11)

## Symptom

`systemctl suspend` (manual or idle-triggered) enters s2idle successfully — freezer, NetworkManager sleep, EC interrupt block/unblock all complete normally — but the system resumes on its own 2-6 seconds later. No lid event, no button press, no `systemctl suspend` reissued. Repeats indefinitely: idle timer re-arms, suspends, wakes again within seconds.

journalctl -k excerpt from one cycle:
```
Aug 12 08:11:27 kernel: PM: suspend entry (s2idle)
Aug 12 08:11:27 kernel: Filesystems sync: 0.005 seconds
Aug 12 08:11:31 kernel: Freezing user space processes
...
Aug 12 08:11:31 kernel: PM: suspend exit
```
(4 seconds total, no intervening wake source logged)

`/sys/power/pm_wakeup_irq` reads "No data available" immediately after resume — no IRQ is attributed to the wake, which rules out a specific device (USB/BT/lid) triggering it.

## amd-s2idle diagnostic output

Ran `amd-s2idle test` (amd-debug-tools) on 7.1.8-1:

```
The system was programmed to sleep for 0:00:10, but woke up prematurely after
0 days 00:00:06. This typically happens when the system was woken up from a
non-timer based source. If you didn't intentionally wake it up, then there
may be a kernel or firmware bug.
```

`gpe0A` counter increased by 13-16 during these failed cycles (e.g. 1765 -> 1781), suggesting some ACPI GPE is firing during the sleep window, though I haven't isolated which specific GPE bit.

Same test on linux-cachyos-lts 6.18.42-1, same hardware, same BIOS: completes the full programmed sleep duration, no premature-wake warning.

## Reproduction

```
sudo pipx run --spec amd-debug-tools amd-s2idle test --count 1 --format txt --report-file /tmp/report.txt
```
on 7.1.8-1 reliably reproduces the premature wake. Booting linux-cachyos-lts and repeating does not.

## Ask

Given this reproduces cleanly on LTS but not on the current 7.1.8-1 branch on identical hardware/firmware, it looks like a regression somewhere in the 7.1.7-7.1.8 patch delta rather than an upstream/firmware issue. Happy to bisect further or pull any additional diagnostics if pointed at what to collect.

Start by reproducing the failure with `amd-s2idle test --count 1 --format txt --report-file /tmp/report.txt` on linux-cachyos 7.1.8-1, then compare it with 7.1.6-1 and linux-cachyos-lts 6.18.42-1. Investigate the 7.1.7–7.1.8 patch delta and collect the ACPI GPE and wakeup diagnostics mentioned in the report. Done means identifying the regression or narrowing it to a specific change, with suspend completing the programmed duration.

Written by the indexing model from the issue text.