vllm-batch for lit5 and firstmistral is supported or not?

[sahel-sh]

--vllm-batched is passed set in LiT5 and FirstMistral examples in the README. But later on we say:

vLLM, SGLang, and TensorRT-LLM backends are only supported for RankZephyr and RankVicuna models.

I think we should have a clear table for which flags are supported for which rerankers. For example, I assume --use_logits and --use_alpha only make sense with the listwise rerankers (or are they only supported with FirstMistral?).

[sahel-sh]

@ronakice would you PTAL?

[XKTZ]

--vllm_batched is used for LiT5 because when LiT5 is integrated vllm is the only batch method so I reused this parameter as a simple "batched" (i.e. not open this will cause model run the doc one by one) If we wanna it more preceise we could probably add another --batched arg probably