Optimized SAC soft update

ml-agents/mlagents/trainers/sac/optimizer_torch.py

         return q1_loss, q2_loss
 
     def soft_update(self, source: nn.Module, target: nn.Module, tau: float) -> None:
-        for source_param, target_param in zip(source.parameters(), target.parameters()):
-            target_param.data.copy_(
-                target_param.data * (1.0 - tau) + source_param.data * tau
-            )
+        with torch.no_grad():
+            for source_param, target_param in zip(
+                source.parameters(), target.parameters()
+            ):
+                target_param.data.mul_(1.0 - tau)
+                torch.add(
+                    target_param.data,
+                    source_param.data,
+                    alpha=tau,
+                    out=target_param.data,
+                )
 
     def sac_value_loss(
         self,