fix(ns-storage): drop the ns_data label when releasing the storage

[gsanchietti]

Summary

The persistent storage is auto-mounted by filesystem label (fstab.ns_data.label=ns_data), but remove-storage only released the old partition when it lived on the OS disk:

if [ "$rom_disk" == "$data_disk" ]; then
    parted "/dev/${data_disk}" rm 3
fi

A storage removed from a secondary drive therefore kept its ns_data label. Once a new storage was created, two partitions matched the label and block mount stacked both onto /mnt/data at boot. ns.storage get-configuration reported the wrong device for the same reason.

Two commits:

1. Release the label. On a secondary drive, clear it with tune2fs -L "" and keep the partition, so its logs stay available for later inspection — the behaviour proposed in the issue. On the OS disk the partition is still deleted, but the partition number now comes from /sys/class/block/<part>/partition instead of the hardcoded 3.
2. Don't abort when nothing is mounted. Pre-existing, found while testing: with no storage configured, lsblk -lno pkname /dev/ exits 32 and set -e killed the script before any teardown ran. ns.factoryreset invokes remove-storage unconditionally, so a factory reset silently skipped the storage teardown entirely.

No new dependencies — parted and tune2fs are already in DEPENDS.

Related issue

#1862

How to test

On a machine with the OS disk plus a second drive:

1. Delete the storage if one exists, then create a new one on the secondary drive.
2. Delete it. Check the label is gone but the partition is not:

   lsblk -lno KNAME,LABEL   # no ns_data
   lsblk /dev/<second-disk> # partition still there
   

3. Create a storage on a partition of the primary drive, then reboot.
4. Only one filesystem must be mounted:

   grep /mnt/data /proc/mounts   # a single line
   storage-status                # ok
   api-cli ns.storage get-configuration   # reports the primary drive partition
   

Reverse cycle (primary → delete → secondary) should leave partition 3 gone from the primary drive's GPT and the secondary partition as the only ns_data.

For the second commit, run /usr/sbin/remove-storage with no storage configured: it must exit 0 and leave dnsmasq, rsyslog, cron and victoria-metrics running.

Verified on a dev VM (OS on vda with a bios_grub partition 128, secondary nvme0n1): all of the above, before and after reboot.

Known limitation

Systems that are already double-mounted are not repaired automatically — data_part resolves to two devices there. They need one manual tune2fs -L "" /dev/<stale-partition>; afterwards this fix keeps them correct.