[Feat] Support routing replay (#6696)

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

[Feat] Support routing replay
same as https://github.com/vllm-project/vllm-ascend/pull/6666
resubmit  because of DOC failure

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

### How was this patch tested?

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

---------

Signed-off-by: liyongwen <1310439159@qq.com>
Signed-off-by: Li-Yongwen <63399187+Li-Yongwen@users.noreply.github.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>

vllm_ascend/ops/fused_moe/fused_moe.py

@@ -26,6 +26,7 @@ from vllm.forward_context import get_forward_context
 from vllm.logger import logger
 from vllm.model_executor.layers.fused_moe.config import FusedMoEConfig
 from vllm.model_executor.layers.fused_moe.layer import FusedMoE, UnquantizedFusedMoEMethod, get_compressed_expert_map
+from vllm.model_executor.layers.fused_moe.routed_experts_capturer import RoutedExpertsCapturer
 from vllm.model_executor.layers.fused_moe.shared_fused_moe import SharedFusedMoE
 
 from vllm_ascend.utils import vllm_version_is
@@ -122,6 +123,13 @@ class AscendUnquantizedFusedMoEMethod(UnquantizedFusedMoEMethod):
             e_score_correction_bias=e_score_correction_bias,
             global_num_experts=global_num_experts,
         )
+        if layer.vllm_config.model_config is not None and layer.vllm_config.model_config.enable_return_routed_experts:
+            capturer = RoutedExpertsCapturer.get_instance()
+            if capturer is not None:
+                capturer.capture(
+                    layer_id=layer.layer_id,
+                    topk_ids=topk_ids,
+                )
 
         if zero_expert_num > 0 and zero_expert_type is not None:
             topk_ids, topk_weights, zero_expert_result = zero_experts_compute(