[Main] [Refactor] Enable MoECommMethod in Eager Mode (#2791)

### What this PR does / why we need it?
1. Replace prepare/finalize operation in fused_moe.py by
moe_comm_method.prepare()/finalize()
2. Replace unified_fused_experts by moe_comm_method.fused_experts() in
fused_moe.py/w8a8_dynamic.py/w4a8_dynamic.py
3. Add calling _select_moe_comm_method in spec-decode proposers.
4. Currently, w4a8_dynamic does not support gatherep, use all2allv
instead.
5. Remove redundant code.
### Does this PR introduce _any_ user-facing change?
AllgatherEP switch is disabled in aclgraph/eager mode, just follow the
rules in modelrunner_v1._select_moe_comm_method()
### How was this patch tested?
e2e & ut


- vLLM version: v0.10.2
- vLLM main:
7f6f2c1182

Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
Co-authored-by: weijinqian0 <12153182+weijinqian0@users.noreply.github.com>

vllm_ascend/quantization/w4a8_dynamic.py

@@ -24,9 +24,7 @@ from vllm.config import get_current_vllm_config
 from vllm.distributed import get_ep_group
 from vllm.forward_context import get_forward_context
 
-from vllm_ascend.ascend_forward_context import FusedMoEState
 from vllm_ascend.distributed.parallel_state import get_mc2_group
-from vllm_ascend.ops.fused_moe import unified_fused_experts_eager
 from vllm_ascend.ops.moe.experts_selector import select_experts
 
 
@@ -275,14 +273,6 @@ class AscendW4A8DynamicFusedMoEMethod:
             e_score_correction_bias=e_score_correction_bias,
             global_num_experts=global_num_experts)
 
-        fused_moe_state = get_forward_context().fused_moe_state
-        shared_gate_up, shared_dequant_scale = None, None
-        if shared_experts is not None and fused_moe_state == FusedMoEState.MC2:
-            share_up_out, _ = shared_experts.gate_up_proj(
-                (quantized_x_for_share, dynamic_scale_for_share))
-            shared_gate_up, shared_dequant_scale = share_up_out[
-                0], share_up_out[1]
-
         # this is a naive implementation for experts load balance so as
         # to avoid accumulating too much tokens on a single rank.
         # currently it is only activated when doing profile runs.
@@ -291,7 +281,8 @@ class AscendW4A8DynamicFusedMoEMethod:
 
         topk_weights = topk_weights.to(x.dtype)
 
-        return unified_fused_experts_eager(
+        moe_comm_method = get_forward_context().moe_comm_method
+        return moe_comm_method.fused_experts(
             hidden_states=x,
             w1=layer.w13_weight,
             w2=layer.w2_weight,
@@ -302,14 +293,13 @@ class AscendW4A8DynamicFusedMoEMethod:
             topk_weights=topk_weights,
             topk_ids=topk_ids,
             row_idx=row_idx,
+            use_int4_w4a8=True,
             expert_map=expert_map,
             log2phy=log2phy,
             global_redundant_expert_num=global_redundant_expert_num,
             shared_experts=shared_experts,
-            shared_gate_up=shared_gate_up,
-            shared_dequant_scale=shared_dequant_scale,
-            mc2_mask=kwargs.get("mc2_mask", None),
-            with_quant=True)
+            quantized_x_for_share=quantized_x_for_share,
+            dynamic_scale_for_share=dynamic_scale_for_share)
 
     def process_scale(self, weight: torch.Tensor, scale, per_group_scale):
         group_num, k, n = weight.shape