Add pad token to tokenizer

Whisper large is missing the pad token, which is otherwise added in the tiny-medium models (e.g. https://huggingface.co/openai/whisper-medium/blob/main/special_tokens_map.json#L125). This PR adds the pad token for the large checkpoint.

special_tokens_map.json

@@ -122,6 +122,7 @@
     "rstrip": false,
     "single_word": false
   },
+  "pad_token": "<|endoftext|>",
   "unk_token": {
     "content": "",
     "lstrip": false,