Fix weight decay bug for `SGD` optimizer.

Weight decay has been considered and caculated in `SGD` operation, so
there's no need to apply weight decay in `SGD` optimizer.

mindspore/nn/optim/sgd.py

@@ -136,7 +136,6 @@ class SGD(Optimizer):
         params = self.parameters
         accum = self.accum
         stat = self.stat
-        gradients = self.decay_weight(gradients)
         gradients = self.scale_grad(gradients)
         lr = self.get_lr()
         if self.is_group_lr: