Trainer - deprecate tokenizer for processing_class (#32385)

* Trainer - deprecate tokenizer for processing_class

* Extend chage across Seq2Seq trainer and docs

* Add tests

* Update to FutureWarning and add deprecation version

docs/source/ja/tasks/token_classification.md

@@ -295,7 +295,7 @@ pip install transformers datasets evaluate seqeval
 ...     args=training_args,
 ...     train_dataset=tokenized_wnut["train"],
 ...     eval_dataset=tokenized_wnut["test"],
-...     tokenizer=tokenizer,
+...     processing_class=tokenizer,
 ...     data_collator=data_collator,
 ...     compute_metrics=compute_metrics,
 ... )