Remove redundant tensor assignment in model runner.

The `assign` operation was unnecessary as new tensors are realized for updated inputs. This simplifies the code and avoids redundant updates, improving clarity and maintainability.

selfdrive/modeld/runners/model_runner.py

@@ -70,8 +70,6 @@ class TinyGradRunner(ModelRunner):
     for k, v in numpy_inputs.items():
       if k not in self.tensor_inputs:
         self.tensor_inputs[k] = Tensor(v, device='NPY').realize()
-      else:
-        self.tensor_inputs[k].assign(v)
 
     return self.tensor_inputs