[Perf][Diffusion] Compile and prewarm VAE decode

Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>

vllm_omni/engine/async_omni_engine.py

@@ -1043,6 +1043,9 @@ class AsyncOmniEngine:
             "diffusion_compile_dynamic": (
                 True if kwargs.get("diffusion_compile_dynamic") is None else kwargs["diffusion_compile_dynamic"]
             ),
+            "diffusion_compile_vae": (
+                True if kwargs.get("diffusion_compile_vae") is None else kwargs["diffusion_compile_vae"]
+            ),
             "fa_deterministic": bool(kwargs.get("fa_deterministic", False)),
             "boundary_ratio": kwargs.get("boundary_ratio", None),
             "flow_shift": kwargs.get("flow_shift", None),