Lock GPU pin with warm CUDA alloc before ultralytics select_device runs

Ultralytics select_device() rewrites CUDA_VISIBLE_DEVICES on first
predict() when the CUDA context is uninitialized: --gpu 1 silently
became a two-GPU cuda:0,1 context that OOMed on GPU0 (Strata's card).
Initializing torch and touching cuda once after setting CVD locks the
pin; both scripts now use device cuda:0 within the pin.

audit_stills.py

@@ -40,8 +40,16 @@ def main():
         os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
         device = "cpu"
     elif args.gpu.lower() != "auto":
+        # Lock the pin with a warm CUDA alloc BEFORE ultralytics runs
+        # select_device(), which would otherwise rewrite
+        # CUDA_VISIBLE_DEVICES and un-pin us onto GPU0.
         os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
-        device = "cuda"
+        import torch
+        if torch.cuda.is_available():
+            torch.zeros(1, device="cuda")
+            device = "cuda:0"
+        else:
+            device = "cpu"
     else:
         device = None