[Bugfix] fix custom op GmmSwigluQuantWeightNzTensorList (#4593)

### What this PR does / why we need it?

1. Fixes the environment path used to locate custom op shared libraries.
2. Uses empty tensor initialization for op outputs instead of
zero-initialization for better efficiency.



- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2

---------

Signed-off-by: QianChenxi <chenxi.qian.cq@outlook.com>

vllm_ascend/utils.py

@@ -247,6 +247,19 @@ def enable_custom_op():
     Ensure that ASCEND_RT_VISIBLE_DEVICES can be dynamically modified before torch.npu.set_device().
     """
     global _CUSTOM_OP_ENABLED
+
+    # set custom ops path
+    CUR_DIR = os.path.dirname(os.path.realpath(__file__))
+    CUSTOM_OPP_PATH = os.path.join(CUR_DIR, "_cann_ops_custom", "vendors",
+                                   "vllm-ascend")
+    if os.path.exists(CUSTOM_OPP_PATH):
+        current_cust_opp_path = os.environ.get("ASCEND_CUSTOM_OPP_PATH", "")
+        if current_cust_opp_path:
+            os.environ[
+                "ASCEND_CUSTOM_OPP_PATH"] = f"{CUSTOM_OPP_PATH}:{current_cust_opp_path}"
+        else:
+            os.environ["ASCEND_CUSTOM_OPP_PATH"] = CUSTOM_OPP_PATH
+
     if _CUSTOM_OP_ENABLED is not None:
         return _CUSTOM_OP_ENABLED
     try: