fix: unskip multimodal_disagg_qwen EPD test (#7523)

Signed-off-by: Krishnan Prashanth <kprashanth@nvidia.com>

tests/serve/test_vllm.py

@@ -386,8 +386,7 @@ vllm_configs = {
         marks=[
             pytest.mark.gpu_1,
             pytest.mark.max_vram_gib(19.4),  # observed peak 17.6 GiB (+10% safety)
-            pytest.mark.post_merge,
-            pytest.mark.skip(reason="DYN-2265"),
+            pytest.mark.pre_merge,
         ],
         model="Qwen/Qwen3-VL-2B-Instruct",
         script_args=["--model", "Qwen/Qwen3-VL-2B-Instruct", "--single-gpu"],