[CustomOp] Register AscendSharedFusedMoE custom op (#2980)

### What this PR does / why we need it?
Register `AscendSharedFusedMoE` custom op.

### Does this PR introduce _any_ user-facing change?

No.

### How was this patch tested?

`DeepSeek-V2-Lite` is a MoE model with shared experts.

Test:

```bash
vllm serve /root/.cache/modelscope/hub/models/deepseek-ai/DeepSeek-V2-Lite \
--trust-remote-code \
--enforce-eager \
--no-enable-prefix-caching \
--gpu-memory-utilization 0.95

curl -X POST http://localhost:8000/v1/chat/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "/root/.cache/modelscope/hub/models/deepseek-ai/DeepSeek-V2-Lite",
        "messages": [
            {"role": "user", "content": "介绍一下联通公司？"}
        ],
        "stream": false,
        "max_tokens": 100
    }'
```

Output:

```bash
中国联合网络通信集团有限公司（简称“中国联通”）于2009年1月6日在原中国网通和原中国联通的基础上合并组建而成，在国内31个省（自治区、直辖市）和境外多个国家和地区设有分支机构，是中国唯一一家在纽约、香港、上海三地同时上市的电信运营企业，连续多年入选“世界500强企业”。\n\n中国联通主要经营固定通信业务，移动通信业务，国内
```


- vLLM version: v0.10.2
- vLLM main:
486c5599e3

---------

Signed-off-by: Shanshan Shen <87969357+shen-shanshan@users.noreply.github.com>
Signed-off-by: shen-shanshan <467638484@qq.com>

vllm_ascend/ops/common_fused_moe.py

@@ -27,6 +27,7 @@ from vllm.model_executor.layers.fused_moe.config import \
     FusedMoEParallelConfig  # isort: skip
 from vllm.model_executor.layers.fused_moe.layer import (
     FusedMoE, UnquantizedFusedMoEMethod, determine_expert_map)
+from vllm.model_executor.layers.shared_fused_moe import SharedFusedMoE
 
 from vllm_ascend.ascend_config import get_ascend_config
 from vllm_ascend.distributed.parallel_state import get_mc2_group
@@ -415,7 +416,7 @@ class AscendFusedMoE(FusedMoE):
         expert_data.copy_(loaded_weight)
 
 
-class AscendSharedFusedMoE(AscendFusedMoE):
+class AscendSharedFusedMoE(SharedFusedMoE, AscendFusedMoE):
 
     def __init__(
         self,
@@ -423,7 +424,7 @@ class AscendSharedFusedMoE(AscendFusedMoE):
         use_overlapped: bool = True,
         **kwargs,
     ):
-        super().__init__(**kwargs)
+        AscendFusedMoE.__init__(self, **kwargs)
         self._shared_experts = shared_experts
         self.use_overlapped = use_overlapped
         self.shared_expert_stream = None
@@ -452,7 +453,8 @@ class AscendSharedFusedMoE(AscendFusedMoE):
             if moe_comm_method_name in {"alltoallcommimpl", "mc2commimpl"}:
                 shared_out = tensor_model_parallel_all_reduce(shared_out)
 
-        fused_out = super().forward(
+        _, fused_out = AscendFusedMoE.forward(
+            self,
             hidden_states=hidden_states,
             router_logits=router_logits,
         )
@@ -461,6 +463,16 @@ class AscendSharedFusedMoE(AscendFusedMoE):
             torch.npu.current_stream().wait_stream(self.shared_expert_stream)
         return shared_out, fused_out
 
+    def forward_impl(self, hidden_states: torch.Tensor,
+                     router_logits: torch.Tensor):
+        shared_output = torch.empty(1)
+        fused_output = AscendFusedMoE.forward_impl(
+            self,
+            hidden_states=hidden_states,
+            router_logits=router_logits,
+        )
+        return shared_output, fused_output
+
 
 UnquantizedFusedMoEMethod.__init__ = unquantized_fused_moe_init_func
 UnquantizedFusedMoEMethod.process_weights_after_loading = process_weights_after_loading