fix: provisioner now extracts archives and discovers disks before conversion

The job submission receives the original filename (e.g. Debian_13_VMG.7z)
but qemu-img convert needs the actual VMDK path inside the archive.
Now reuse extract_if_needed() + discover_disk() from converter module
to locate the real disk image before conversion.

backend/provisioner.py

@@ -24,10 +24,10 @@ from models import (
     JobStatus,
     DiskType,
 )
+from converter import extract_if_needed, discover_disk
 
 logger = logging.getLogger("backend.provisioner")
 
-STAGING_IN = Path("/mnt/converter/in")
 STAGING_OUT = Path("/mnt/converter/out")
 
 # In-memory job tracking — survives as long as the process runs
@@ -164,13 +164,15 @@ def _process_job(job_id: str, req: JobSubmissionRequest):
             output_path = out_dir / f"vm-{vmid}-disk-{idx}.qcow2"
 
             if spec.disk_type == DiskType.IMAGE_FILE:
-                source_path = STAGING_IN / spec.source_filename
+                # Extract archive if needed, then locate the actual disk image
-                if not source_path.exists():
+                source = extract_if_needed(spec.source_filename)
-                    raise FileNotFoundError(f"Source file not found: {source_path}")
+                disk = discover_disk(source)
+                source_path = disk.path
+                logger.info("Disk %d: source=%s format=%s size=%.1f GiB",
+                            idx, source_path, disk.format, disk.size_gb)
 
-                logger.info("Converting disk %d: %s → %s", idx, source_path, output_path)
                 _run(["qemu-img", "convert",
-                      "-f", spec.format.value if spec.format else "vmdk",
+                      "-f", disk.format,
                       "-O", "qcow2",
                       str(source_path), str(output_path)],
                      timeout=7200)