[Feature]EPLB:Adapt DispatchGmmCombineDecode operator to eplb tensor list and expert token numbers (#5552)

#### What this PR does / why we need it?
This PR adapt DispatchGmmCombineDecode operator to eplb tensor list and
expert token numbers.

This operator support gmm1, gmm2, gmm1Scale and gmm2Scale in format of
list.
This operator support couting how many token each local expert recieves
by expertTokensNum .


- vLLM version: v0.13.0
- vLLM main:
7157596103

More info about this operator, please refer to RFC: issue
https://github.com/vllm-project/vllm-ascend/issues/5476

vllm_ascend/quantization/w8a8_dynamic.py

@@ -254,7 +254,8 @@ class AscendW8A8DynamicFusedMoEMethod:
             w1 = layer.w13_weight_list
             w1_scale = layer.w13_weight_scale_fp32_list
             w2 = layer.w2_weight_list
-            w2_scale = layer.w2_weight_scale_list
+            w2_scale = layer.w2_weight_scale_fp32_list \
+                if w2_weight_scale_fp32_flag else layer.w2_weight_scale_list
         else:
             w1 = [layer.w13_weight]
             w1_scale = [layer.w13_weight_scale_fp32]
@@ -333,11 +334,16 @@ class AscendW8A8DynamicFusedMoEMethod:
                 weight.clone()
                 for weight in layer.w2_weight_scale.data.unbind(dim=0)
             ]
+            layer.w2_weight_scale_fp32_list = [
+                weight.clone()
+                for weight in layer.w2_weight_scale_fp32.data.unbind(dim=0)
+            ]
             del layer.w13_weight
             del layer.w2_weight
             del layer.w13_weight_scale
             del layer.w13_weight_scale_fp32
             del layer.w2_weight_scale
+            del layer.w2_weight_scale_fp32
             torch.npu.empty_cache()