Fix encrypted ZFS persistence issue

Problem: VMs with encrypted ZFS storage were reformatting the disk on
every reboot, even with KMS or local key provider enabled.

Root cause: The is_disk_initialized() check required both LUKS header
AND zpool detection for encrypted ZFS (line 1057-1058). However, zpool
import is attempted on the raw encrypted device (/dev/vdb1) before LUKS
is opened, which always fails since the ZFS pool is inside the LUKS
container.

Solution: For any encrypted filesystem (ext4 or ZFS), only check for
LUKS header presence, since the actual filesystem can only be detected
after opening LUKS. This aligns encrypted ZFS handling with encrypted
ext4, which was already using this approach.

Testing: Verified that ext4 encrypted storage persists correctly with
KMS key provider. ZFS testing pending after build completes.

Author: kvinwang

dstack-util/src/system_setup.rs

@@ -1053,11 +1053,12 @@ impl<'a> Stage0<'a> {
             }
         };
 
-        // For encrypted ZFS, need both LUKS header AND zpool to exist
-        let initialized = if opts.storage_encrypted && opts.storage_fs == FsType::Zfs {
-            has_luks && has_fs
+        // For encrypted filesystems, we can only detect the filesystem after LUKS is opened
+        // So we rely on LUKS header presence as the indicator for both ext4 and ZFS
+        let initialized = if opts.storage_encrypted {
+            has_luks
         } else {
-            has_luks || has_fs
+            has_fs
         };
 
         if !initialized {