Processor: prevent duplicated tokens

When using text-only LLMs, the chat template is expected to take care of
adding the required special tokens, such as bos. Hence, tokenization
must not include special tokens.

The same contract should be honored for multimodal processors.

src/transformers/processing_utils.py

@@ -1246,6 +1246,7 @@ class ProcessorMixin(PushToHubMixin):
                 text=prompt,
                 images=images if images else None,
                 videos=videos if videos else None,
+                add_special_tokens=False,
                 **kwargs,
             )
             if return_dict: