Mark qwen2_5_vl as xfail (#30283)

Signed-off-by: Yanan Cao <gmagogsfm@gmail.com>

tests/compile/fullgraph/test_multimodal_compile.py

@@ -17,6 +17,7 @@ def test_compile():
 # forked needed to workaround https://github.com/vllm-project/vllm/issues/21073
 @pytest.mark.forked
 @pytest.mark.skipif(not current_platform.is_cuda(), reason="Skip if not cuda")
+@pytest.mark.xfail
 def test_qwen2_5_vl_compilation(vllm_runner, monkeypatch):
     """Test that Qwen2.5-VL vision submodules are compiled.