Arm backend: Avoid tensor copy warning in quant config (#20687)

The scale product is already a tensor, so wrapping it with torch.tensor
emits a PyTorch copy-construction warning. Use it directly and cast it
to float32 instead.

Signed-off-by: Sebastian Larsson <sebastian.larsson@arm.com>

Author: Sebastian-Larsson

backends/arm/quantizer/quantization_config.py

@@ -189,9 +189,9 @@ def _derive_qparams_fn(
             weight_obs_or_fq = obs_or_fqs[1]
             act_scale, _ = act_obs_or_fq.calculate_qparams()
             weight_scale, _ = weight_obs_or_fq.calculate_qparams()
-            return torch.tensor(act_scale * weight_scale).to(
-                torch.float32
-            ), torch.full_like(weight_scale, fill_value=0, dtype=torch.int32)
+            return (act_scale * weight_scale).to(torch.float32), torch.full_like(
+                weight_scale, fill_value=0, dtype=torch.int32
+            )
 
         if node.target in [
             torch.ops.aten.conv1d.default,