Fix crash when using vLLM without structured generation

[cg123]

When using the vLLM backend, currently structured generation must be set or the task will crash with the following error:

 "/usr/local/lib/python3.10/dist-packages/vllm/model_executor/layers/logits_processor.py", line 106, in _apply_logits_processors
    logits_row = logits_processor(token_ids, logits_row)
TypeError: 'NoneType' object is not callable

This is because [self._logits_processor] is passed to vLLM as the list of logit processors, even when self._logits_processor is None. vLLM isn't picky and happily calls that None.

Just a one-line tweak to make things work again.

[alvarobartt]

Thanks for the quick fix @cg123! We'll release distilabel v1.1.1 tomorrow with this, thanks!