[4/n]decouple quantization implementation from vLLM dependency (#9191)

Co-authored-by: AniZpZ <aniz1905@gmail.com>
Co-authored-by: Yineng Zhang <me@zhyncs.com>

.github/workflows/vllm-dependency-test.yml

@@ -30,14 +30,9 @@ jobs:
       - name: Install dependencies
         run: |
           bash scripts/ci/ci_install_dependency.sh
-          pip install "vllm==0.10.0"
-          pip install "openai==1.99.1"
           pip install "bitsandbytes>=0.44.0"
 
-          # NOTE: The latest sgl-kernel depends on torch 2.8.0 but the latest vllm depends on torch 2.7.0
-          # so they are not compatible. Here we install the old sgl-kernel to make the test pass.
-          # TODO: remove this once vllm supports torch 2.8.0.
-          pip install "sgl-kernel==0.2.9"
+          pip install "sgl-kernel==0.3.5"
 
       - name: Run vLLM dependency tests
         timeout-minutes: 60