0.32.1: backup/restore fixes (audit findings, all paths live-verified)

- backup_filename() crashed with NameError (bare now()) since 0.8.0 —
  broke every scheduled backup at the upload step, agent backup download
  and the central remote-backup/push endpoints. The local manual path
  worked only because the router had its own copy (now an alias).
- restore: the manifest stack_id from an uploaded backup is now slugified
  too — a crafted '../../...' id could previously escape STACKS_DIR.
- create_backup no longer starts a previously-stopped stack (stop/restart
  only when the stack was actually running).
- overwrite-restore wipes the existing volume contents before extracting,
  so files created since the backup no longer survive underneath it.

Verified end-to-end: full/config backup contents (compose, .env, .secrets,
bind dirs, extras, volume tars), delete→restore round-trip incl. volume
data, rename restore with volume re-prefixing, 409 conflict + overwrite,
traversal guard, scheduled run + retention prune + restore-from against
real MinIO, and the complete remote-agent cycle (download/push/restore).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

backend/routers/backups.py

@@ -29,10 +29,7 @@ def _ip(request: Request) -> str:
     return request.client.host if request.client else "unknown"
 
 
-def _backup_filename(stack_id: str, include_volumes: bool) -> str:
-    date = compose_service.now().strftime("%Y%m%d-%H%M%S")
-    suffix = "full" if include_volumes else "config"
-    return f"backup-{stack_id}-{suffix}-{date}.tar.gz"
+_backup_filename = backup_service.backup_filename
 
 
 @router.get("/{stack_id}/backup")