Use YOLO-NAS model for TensorRT-compatible ONNX inference

The YOLOv9m model fails with CUDA graph capture on the tensorrt image.
Try YOLO-NAS-S which has a different architecture that may be fully
partitionable to the CUDA execution provider.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

argocd/manifests/frigate/configmap.yaml

@@ -59,6 +59,15 @@ data:
       onnx:
         type: onnx
 
+    model:
+      model_type: yolonas
+      width: 320
+      height: 320
+      input_tensor: nchw
+      input_dtype: float
+      path: /media/frigate/models/yolo_nas_s.onnx
+      labelmap_path: /labelmap/coco-80.txt
+
     record:
       enabled: true
       continuous: