Fix PPOTrainer argument train_dataset -> dataset (#1321)

Both the argument's name as well as the value need to be renamed.
Otherwise we get both

NameError: name 'train_dataset' is not defined

and

TypeError: PPOTrainer.__init__() got an unexpected keyword argument 'train_dataset'

docs/source/ppo_trainer.mdx

@@ -90,7 +90,7 @@ from trl import PPOTrainer
 ppo_trainer = PPOTrainer(
     model=model,
     config=config,
-    train_dataset=train_dataset,
+    dataset=dataset,
     tokenizer=tokenizer,
 )
 ```