fix(tokenizer): set `mode_max_length=4096`

Thanks to

@ctranslate2-4you

tokenizer_config.json

@@ -40,7 +40,7 @@
   "clean_up_tokenization_spaces": true,
   "eos_token": "<|endoftext|>",
   "tokenizer_class": "GPT2TokenizerFast",
-  "model_max_length": 2048,
+  "model_max_length": 4096,
   "pad_token": "<|endoftext|>",
   "unk_token": "<|endoftext|>"
 }