[Bugfix][DispatchFFNCombine] resolve vec error caused by unaligned UB access (#6707)

### What this PR does / why we need it?
1. Fix a vec error caused by unaligned UB accesss in the
DispatchFFNCombine;
2. Fix expert_token_nums tensor defined on host instead of NPU in
moe_comm_method.py
3. Fix multi-core copy issue of expert_token_nums in dispatchffnCombine
op (single aiv copy is sufficient)

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

No, this PR does not introduce any user-facing changes. The fix only
addresses internal memory access logic and does not modify any public
APIs, interfaces, or user-visible behaviors.

### How was this patch tested?

`export VLLM_ASCEND_ENABLE_FUSED_MC2=1`

vLLM version: v0.15.0

- vLLM version: v0.15.0
- vLLM main:
9562912cea

Signed-off-by: xulei_ict <xulei292@huawei.com>
Co-authored-by: xulei_ict <xulei292@huawei.com>

vllm_ascend/ops/fused_moe/moe_comm_method.py

@@ -277,6 +277,13 @@ class FusedMC2CommImpl(MoECommMethod):
     Communication and Computation parallelism on Ascend devices.
     """
 
+    def __init__(self, moe_config):
+        super().__init__(moe_config)
+        if envs_ascend.VLLM_ASCEND_ENABLE_FUSED_MC2 == 1:
+            self.expert_token_nums = torch.zeros([self.moe_config.num_local_experts], dtype=torch.int32, device="npu")
+        else:
+            self.expert_token_nums = None
+
     def _get_token_dispatcher(self):
         return TokenDispatcherWithMC2()
 
@@ -325,7 +332,6 @@ class FusedMC2CommImpl(MoECommMethod):
         expert_tokens = None
         if envs_ascend.VLLM_ASCEND_ENABLE_FUSED_MC2 == 1:
             out = torch.empty_like(hidden_states)
-            expert_token_nums = torch.zeros([self.moe_config.num_local_experts], dtype=torch.int32)
             torch.ops._C_ascend.dispatch_ffn_combine(  # type: ignore
                 x=hidden_states,
                 weight1=w1,
@@ -337,9 +343,9 @@ class FusedMC2CommImpl(MoECommMethod):
                 group=self.token_dispatcher.moe_all_to_all_group_name,
                 max_output_size=65536,
                 out=out,
-                expert_token_nums=expert_token_nums,
+                expert_token_nums=self.expert_token_nums,
             )
-            expert_tokens = expert_token_nums
+            expert_tokens = self.expert_token_nums
         elif envs_ascend.VLLM_ASCEND_ENABLE_FUSED_MC2 == 2:
             assert expert_map is not None, "expert_map cannot be None."
             group_list_type = 1