feat: pp mtp加入零消耗调度，加入环境变量VLLM_USE_ZERO_MTP，默认打开

vllm/envs.py

@@ -182,6 +182,7 @@ if TYPE_CHECKING:
     VLLM_USE_LIGHTOP_FILL_MOE_ALIGN: bool = False
     USE_FUSED_CUSTOM_ALL_REDUCE_RMS_QUANT: bool = False
     VLLM_USE_PP_BALANCE: bool = False
+    VLLM_USE_ZERO_MTP: bool = False
     VLLM_USE_CUDA_GRAPH_SIZES: bool = False
 
 def get_default_cache_root():
@@ -1186,6 +1187,10 @@ environment_variables: dict[str, Callable[[], Any]] = {
         lambda: (os.getenv('VLLM_USE_PP_BALANCE', '1').lower() in
                  ("true", "1")),
 
+    "VLLM_USE_ZERO_MTP":
+        lambda: (os.getenv('VLLM_USE_ZERO_MTP', '1').lower() in
+                 ("true", "1")),
+
     # vllm will use 1-18... (not only 1 2 4 8 16)
     "VLLM_USE_CUDA_GRAPH_SIZES":
         lambda: (os.getenv('VLLM_USE_CUDA_GRAPH_SIZES', 'False').lower() in