Support int `think_end_token` for vllm backend

[fxmarty-amd]

Hi,

As per title.

Transformers backend supports int think_end_token:

lm-evaluation-harness/lm_eval/models/huggingface.py

Line 105 in 29a0765

think_end_token: str | int | None = None,

But vllm backend does not support it:

lm-evaluation-harness/lm_eval/models/vllm_causallms.py

Line 148 in 29a0765

think_end_token: Optional[str] = None,

Thanks!