feat/igpu-ov: OpenVINO iGPU detector (--gpu igpu) on Intel Level Zero

- ov_detector.OVFaceDetector: OV inference + ultralytics letterbox(1280 sq)
  + greedy NMS; model.predict()-compatible shim, process_video() unchanged
- anime_facesaver.py: --gpu igpu branch; CUDA path untouched (moved to else)
- CUDA path untouched; main() unchanged
- validate_ov.py: cross-check vs torch CPU baseline
  measured 2026-10-20 i3-N305: 59/59 matched, mean IoU 0.987,
  only_base=0 only_ov=0; 5.0s/img GPU vs 14.5s/img CPU OpenVINO-CPU vs
  torch-CPU baseline (25 real stills, yolov8x6_animeface)

validate_ov.py

@@ -0,0 +1,41 @@
+"""Cross-validate OVFaceDetector vs ultralytics CPU baseline on real stills."""
+import glob, sys, time
+import numpy as np, cv2
+
+imgs = sorted(glob.glob('/home/alex/Documents/ov-bench/imgs/*.png'))
+sys.path.insert(0, '.')
+from ov_detector import OVFaceDetector
+
+# baseline: ultralytics torch CPU (deterministic on CPU)
+from ultralytics import YOLO
+base = YOLO('models/yolov8x6_animeface.pt')
+
+ovd = OVFaceDetector('models/yolov8x6_animeface.onnx', device='GPU', imgsz=1280)
+
+def iou(a, b):
+    xx1, yy1 = max(a[0], b[0]), max(a[1], b[1])
+    xx2, yy2 = min(a[2], b[2]), min(a[3], b[3])
+    inter = max(0, xx2-xx1) * max(0, yy2-yy1)
+    ua = (a[2]-a[0])*(a[3]-a[1]) + (b[2]-b[0])*(b[3]-b[1]) - inter
+    return inter / max(ua, 1e-9)
+
+ious, n_match, n_only_base, n_only_ov = [], 0, 0, 0
+t0=time.time()
+for p in imgs:
+    fr = cv2.imread(p)
+    rb = base.predict(fr, conf=0.25, imgsz=1280, device='cpu', verbose=False)[0]
+    ro = ovd.predict(fr, conf=0.25, imgsz=1280, verbose=False)[0]
+    bb = rb.boxes.xyxy.cpu().numpy() if rb.boxes is not None else np.zeros((0,4))
+    bo = ro.boxes.xyxy.numpy()
+    used = set()
+    for box in bb:
+        best, bi = 0, -1
+        for j, ob in enumerate(bo):
+            v = iou(box, ob)
+            if v > best: best, bi = v, j
+        if best > 0.5:
+            ious.append(best); n_match += 1; used.add(bi)
+        else: n_only_base += 1
+    n_only_ov += len(bo) - len(used)
+dt = time.time()-t0
+print(f"imgs={len(imgs)} matched={n_match} (mean IoU {np.mean(ious):.3f}) only_base={n_only_base} only_ov={n_only_ov} [{dt:.0f}s total]")