[CI/Build] Fix amd model executor test

[zhewenl]

Purpose

Fix Model Executor Test running on AMD CI:(example)
We need to skip fastsafetensors tests, since it's only supporting CUDA platform.

Test Plan

CI: https://buildkite.com/vllm/amd-ci/builds/635

pytest -v -s model_executor
...
================================================================== warnings summary ==================================================================
<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyPacked has no __module__ attribute

<frozen importlib._bootstrap>:488
  <frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyObject has no __module__ attribute

model_executor/model_loader/tensorizer_loader/test_tensorizer.py:281
  /data/users/zhewenli/gitrepos/vllm-fork/tests/model_executor/model_loader/tensorizer_loader/test_tensorizer.py:281: PytestUnknownMarkWarning: Unknown pytest.mark.flaky - is this a typo?  You can register custom marks to avoid this warning - for details, see https://docs.pytest.org/en/stable/how-to/mark.html
    @pytest.mark.flaky(reruns=3)

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! KeyboardInterrupt !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
/home/zhewenli/uv_env/vllm-fork/lib64/python3.12/site-packages/urllib3/response.py:429: KeyboardInterrupt
(to show a full traceback on KeyboardInterrupt use --full-trace)
=============================================== 13 passed, 2 skipped, 3 warnings in 1592.21s (0:26:32) ===============================================

[yeqcharlotte]

could you create issues to keep track of this?

[zhewenl]

Tracking in #27624