cmdlib.sh: keep trying to remount after killall hack

Michael and I were debugging an issue where `cosa build` would still
break on:

```
+ killall rofiles-fuse
+ /sbin/fstrim -v /home/jenkins/agent/workspace/bootupd_PR-706/cache
/home/jenkins/agent/workspace/bootupd_PR-706/cache: 17.8 GiB (19084976128 bytes) trimmed
+ mount -o remount,ro /home/jenkins/agent/workspace/bootupd_PR-706/cache
mount: /home/jenkins/agent/workspace/bootupd_PR-706/cache: mount point is busy.
```

which I think is because the rofiles-fuse processes haven't fully been
cleaned up yet. Let's just spam retry to remount the cache, but notify
the user when this is happening for more visibility.

We should be able to revert this once we fix
#3848

Author: jlebon

src/cmdlib.sh

@@ -762,7 +762,10 @@ if [ -n "\${cachedev}" ]; then
     # XXX: brutal workaround for https://github.com/coreos/coreos-assembler/issues/3848
     killall rofiles-fuse || :
     /sbin/fstrim -v ${workdir}/cache
-    mount -o remount,ro ${workdir}/cache
+    while ! mount -o remount,ro ${workdir}/cache; do
+        echo "failed to remount cache ro; retrying..." |& tee /dev/virtio-ports/cosa-cmdout
+        sleep 1
+    done
     fsfreeze -f ${workdir}/cache
     fsfreeze -u ${workdir}/cache
     umount ${workdir}/cache