feature(wqj): add vllm rlhf collector

[wqj2004]

Overview

The whole RLHF training pipeline

Goal

In this PR, we need to finish the collector part in the above figure and add its unittest.

TODO

• add vllm inferencer for LLM/VLM
• scale up vllm inferencer in multi-GPUs
• add several datasets definition @PaParaZz1 feature(nyz): add rlhf dataset #854
• survey proper VQA dataset for RLHF training
• add the final collector and test it on the above dataset
• add tutorial and API doc
• (optional) add search tools @PaParaZz1

[PaParaZz1]

add python typing lint

[PaParaZz1]

use enum class to control this field rather than the naive string