[CI/Build] Split diffusion distributed nightly tests by L4 and H100 (#5695)

Signed-off-by: wangyu <410167048@qq.com>

tests/diffusion/distributed/test_vae_decode_parallelism.py

@@ -39,7 +39,7 @@ MODEL_CASES = [
             "needs_image": False,
             "is_moe": False,
             "mean_threshold": 3e-2,
-            "p99_threshold": 3e-2,
+            "p99_threshold": 4e-2,
         },
         id="wan22_t2v_a14b",
     ),
@@ -146,7 +146,7 @@ def _run_generate(
 @pytest.mark.full_model
 @pytest.mark.diffusion
 @pytest.mark.parallel
-@hardware_test(res={"cuda": "L4", "rocm": "MI325"}, num_cards={"cuda": 4, "rocm": 2})
+@hardware_test(res={"cuda": "H100"}, num_cards={"cuda": 4})
 @pytest.mark.parametrize("model_case", MODEL_CASES)
 def test_vae_patch_parallel_tp2(model_case: dict[str, Any], tmp_path: Path):
     if current_omni_platform.is_npu():