[internal] Add rawfile support

[duckhawk]

Description

This PR adds support for RawFile volumes — local file-based volumes mounted as loop devices. They provide a simpler alternative to LVM for local storage when LVM is not available or not desired.

Key changes:

1. New rawfile package (images/sds-local-volume-csi/pkg/rawfile/) — handles raw-file/loop-device management:

   • Create/delete raw files (regular or sparse)
   • Attach/detach loop devices (losetup)
   • Online resize (file + loop device + filesystem)
   • List/inspect local volumes; per-volume locking against concurrent operations

2. CSI driver updates (images/sds-local-volume-csi/driver/):

   • CreateVolume: validates topology and pins each volume to a single node taken from accessibility_requirements (no implicit fallback to the controller's host)
   • NodeStageVolume: fetches the PV before file creation, persists .reclaim and .sparse on-disk markers, adds the storage.deckhouse.io/rawfile-pv-protection finalizer for Delete policy, then creates the file and attaches the loop device
   • NodeUnstageVolume / NodeExpandVolume: detach + clean unmount; online resize
   • DeleteVolume: defers actual deletion to the node-side cleanup goroutine (driven by on-disk markers + finalizer)

3. LocalStorageClass CRD extension (api/v1alpha1/, crds/):

   • New rawFile spec with sparse (immutable) and nodes (mutable)
   • CEL XValidation for mutual exclusivity with lvm, immutability of backend switching and rawFile.sparse, mutability of rawFile.nodes

4. Controller updates (images/controller/):

   • StorageClass parameters for the RawFile type (type=rawfile, rawfile-sparse, rawfile-nodes)
   • rawFile.nodes propagated to StorageClass.AllowedTopologies (key topology.sds-local-volume-csi/node) so both the scheduler (WFFC) and external-provisioner (Immediate) honor the constraint
   • Drift detection in hasSCDiff for rawFile.sparse and rawFile.nodes triggers SC recreation

5. Webhook validation (images/webhooks/):

   • Separate validation for RawFile and LVM configurations
   • Rejects empty backend, non-empty/duplicate node names, and missing Node objects (fail-open on transient API errors)

6. Cleanup mechanism (node-side):

   • On-disk .reclaim / .sparse markers stored next to each raw file
   • Background goroutine in the CSI node pod deletes orphaned RawFile volumes (respects grace period; honors ReclaimPolicy: Retain via the marker)
   • Finalizer-based PV protection (storage.deckhouse.io/rawfile-pv-protection)

7. ModuleConfig (openapi/config-values.yaml):

   • New rawFileDefaultDataDir parameter (default: /var/lib/sds-local-volume/rawfile)
   • Wired through the CSI node DaemonSet as RAWFILE_DEFAULT_DATA_DIR and a Bidirectional-propagated hostPath mount

Why do we need it, and what problem does it solve?

Problem. Some environments cannot use LVM due to:

• Lack of available block devices for LVM
• Administrative restrictions on LVM usage
• Need for simpler storage provisioning without LVM overhead
• Testing and development environments where LVM setup is impractical

Solution. RawFile volumes provide:

• File-based storage backed by loop devices on top of any local filesystem
• No LVM dependency
• Simple setup — just specify a directory path (per-node optional via rawFile.nodes)
• Full CSI compatibility (mount, expand, delete)
• Online resize without pod restart

What is the expected result?

After applying these changes:

1. New LocalStorageClass with RawFile:

apiVersion: storage.deckhouse.io/v1alpha1
kind: LocalStorageClass
metadata:
  name: rawfile-storage
spec:
  rawFile:
    sparse: false
    # Optional. If set, restricts provisioning to these nodes via
    # StorageClass.AllowedTopologies.
    # nodes:
    #   - name: node-1
    #   - name: node-2
  reclaimPolicy: Delete
  volumeBindingMode: WaitForFirstConsumer

2. PVC creation works:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-rawfile-pvc
spec:
  storageClassName: rawfile-storage
  accessModes: [ReadWriteOnce]
  resources:
    requests:
      storage: 1Gi

3. Pod scheduling works — no volume node affinity conflict errors (single-node pinning via WFFC + topology label).
4. Online resize works — edit PVC size, the filesystem expands without pod restart.
5. Delete reclaim — files are cleaned up by the node-side cleanup loop after PV finalizer is removed.
6. Retain reclaim — files are preserved (.reclaim=Retain marker).

Checklist

• The code is covered by unit tests.
• e2e tests passed.
• Documentation updated according to the changes.
• Changes were tested in the Kubernetes cluster manually.