lctl snapshot_* works only for ZFS targets. md-kmec (https://github.com/scopedog/mdraid-super, which pins md-kmec and its mdadm) is an md RAID personality (raidkm). Native snapshots are a new md-kmec feature: crash-safe, read-only snapshots of a whole array, managed with mdadm --snapshot from the mdadm pinned there (mdraid-super 895933c: md-kmec b64f85d, mdadm 5bf8fec6). An ldiskfs target that is a whole md-kmec array can use these for Lustre snapshots: mdadm --snapshot --create freezes the filesystem on the array while it takes the snapshot, so a snapshot taken under the write barrier is a clean image.

Changes

• ldev.conf accepts md:/dev/<array>. A snapshot piece is the native snapshot of the array; its read-only device comes from mdadm --snapshot --list, and the lustre:* ZFS properties are kept as array snapshot attributes.
• New mount.lustre option fsname=<name>, allowed only with rdonly_dev, so a snapshot starts under its own fsname with nothing written into it.
• ldev.conf also accepts lvm:/dev/<vg>/<lv> for an LVM thin LV; the md: backend uses the same code. lctl snapshot_create warns when the thin pool geometry does not fit the md row.
• Needs rdonly_dev on ldiskfs (LU-15873).
• sanity-lsnapshot runs on ldiskfs when every target is an md-kmec array or an LVM thin LV; new test_7 (destroy the middle of three snapshots). New conf-sanity test_176 for fsname=.

Testing
Single node, RHEL 9.7 (kernel 5.14.0-611.55.1), md-kmec from mdraid-super 895933c, raidkm 8+2 arrays, 128K chunk. conf-sanity 176 and sanity-lsnapshot 0, 1a, 1b, 2, 3a, 3b, 7 pass with combined and with separate MGS, both with native md-kmec snapshots (one array per target) and with LVM thin LVs in one pool on an md-kmec array. test_4 is excluded: it fails with a separate MGS on every backend, including ZFS, which CI never runs.
With an earlier md-kmec build (same snapshot freeze path), snapshots survived an array stop and assemble and a hard reset (journal replay), and mounted with the data from create time.
CI has no md-kmec arrays, so CI only checks that the ZFS path is unchanged.