Feat: Vulkan/RADV-Engine + vocab-gepruefte Spec-Drafts + Provisioning/Sync

Engine-Cutover ROCm/HIP -> Vulkan/RADV (gfx1151): +12-22% tg auf MoE (llama-bench
verifiziert, fast 53->65 t/s). ROCm-Build bleibt als Rollback unter /opt/llamacpp.

- Backend: vocab-aware Speculative Decoding. services/gguf_meta.py liest den
  Tokenizer-Fingerprint (model/pre/n_vocab) direkt aus dem GGUF-Header (ohne Modell-Load);
  register_model + migrate_config haengen nur VOCAB-KOMPATIBLE Drafts an (inkl. --spec-type,
  das in dieser llama.cpp-Generation noetig ist). Neue Endpoints /api/models/drafts + /{id}/draft.
- Frontend: idiotensichere Spec-Draft-UI (SpecDraftModal) - nur kompatible Drafts waehlbar,
  inkompatible gesperrt mit Begruendung; SPEC/SPEC?-Badge nach echtem Aktiv-Status; Rolle in AddModel.
- maintenance.py: Engine-Update-Quelle -> ggml-org/llama.cpp (Build-Nummer-Vergleich),
  ENGINE_PATH=/opt/llamacpp-vulkan.
- Startup-Warmup der brains (deploy/warmup.sh, self-detaching ExecStartPost) + deploy/provision-engine.sh.
- Cleanup: tote LiteLLM gateway/config.yaml + alle Referenzen (config.py/backup.py/backup.sh) entfernt;
  README + docs/memory aktualisiert.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/services/backup.py

@@ -7,7 +7,7 @@ import shutil
 import time
 from pathlib import Path
 
-from config import CONFIG_PATH, GATEWAY_CONFIG_PATH, MEMORY_DB, MODELS_DIR
+from config import CONFIG_PATH, MEMORY_DB, MODELS_DIR
 
 BACKUP_DIR = Path(MODELS_DIR) / "mc2-backups"
 RETAIN = 7
@@ -31,7 +31,7 @@ def backup_now() -> dict:
     dst.mkdir(parents=True, exist_ok=True)
     saved = []
     for src in (MEMORY_DB, Path(str(MEMORY_DB) + "-wal"), Path(str(MEMORY_DB) + "-shm"),
-                CONFIG_PATH, GATEWAY_CONFIG_PATH):
+                CONFIG_PATH):
         if (name := _safe_copy(src, dst)):
             saved.append(name)
     # Aufräumen: nur die letzten RETAIN Snapshots behalten.