Don't restart a zram device that exists but is swapped off

The resize guard reads the Used column for /dev/zram0 out of /proc/swaps
and treats a missing row as an empty device, which is right for a device
that doesn't exist yet: the restart is what brings it up against the
config daemon-reload just generated. A device that exists and is merely
swapped off reads the same, and there the restart resets it first, which
returns EBUSY for as long as anything still holds it open. That leaves a
bare "Job failed. See 'journalctl -xe' for details." in the migration
output and falls through to asking for the reboot that would have
resized it anyway.

Tell the two apart by whether /sys/block/zram0/disksize is there at all.

The test modelled an absent device as a blank disksize file, which no
longer stands in for one, so it removes the file instead.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

migrations/1785094500.sh

@@ -30,7 +30,16 @@ if sudo systemctl daemon-reload; then
   # pressure keeps its old size until the next boot.
   zram_used=$(awk '$1 == "/dev/zram0" {print $4}' "$swaps")
 
-  if [[ ${zram_used:-0} == 0 ]] && sudo systemctl restart dev-zram0.swap; then
+  # No row at all means the device is not swap right now, and that covers two
+  # unlike situations. A device that doesn't exist yet is the one worth acting
+  # on: the restart brings it up against the config daemon-reload just picked
+  # up. A device that exists but is swapped off is not, because the restart
+  # resets it first, and reset returns EBUSY for as long as anything still
+  # holds it open. That leaves a "Job failed" from systemd in the migration
+  # output and still ends up asking for the reboot, so go straight there.
+  if [[ -n $zram_used || ! -e $zram_disksize ]] &&
+    [[ ${zram_used:-0} == 0 ]] &&
+    sudo systemctl restart dev-zram0.swap; then
     exit 0
   fi
 fi