Fix incorrect code-snippet in optimizers doc (#7598)

`training_step(...)` should take `self` as the first argument. It's a simple but necessary fix.

docs/source/common/optimizers.rst

@@ -49,7 +49,7 @@ Here is a minimal example of manual optimization.
  # Important: This property activates manual optimization.
  self.automatic_optimization = False
 
- def training_step(batch, batch_idx):
+ def training_step(self, batch, batch_idx):
  opt = self.optimizers()
  opt.zero_grad()
  loss = self.compute_loss(batch)