Fix: correct wrong vLLM rerank model (#5399)

### What problem does this PR solve?

Correct wrong vLLM rerank model #4316 

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)

rag/llm/__init__.py

     "NVIDIA": NvidiaRerank,
     "LM-Studio": LmStudioRerank,
     "OpenAI-API-Compatible": OpenAI_APIRerank,
-    "VLLM": OpenAI_APIRerank,
+    "VLLM": CoHereRerank,
     "Cohere": CoHereRerank,
     "TogetherAI": TogetherAIRerank,
     "SILICONFLOW": SILICONFLOWRerank,