[Refactor] Move MM data parsing outside processor (#33408)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/model_executor/models/whisper.py

@@ -705,7 +705,7 @@ class WhisperMultiModalProcessor(EncDecMultiModalProcessor[WhisperProcessingInfo
     def create_encoder_prompt(
         self,
         prompt: str | list[int],
-        mm_data: MultiModalDataDict,
+        mm_items: MultiModalDataItems,
     ) -> str | list[int]:
         # Strictly speaking, whisper encoder only accept audio features.
         # We create a dummy encoder prompt here which will be padded to