[Bugfix]Fix of Pooling Code and Update of Pooling Usage Guide (#6126)

### What this PR does / why we need it?
Fix of Pooling Code and Update of Pooling Usage Guide
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?
pr:[[Bugfix]Fixed precision issues caused by pooled request
pooling](https://github.com/vllm-project/vllm-ascend/pull/6049)
readyhttps://github.com/vllm-project/vllm-ascend/pull/6049
read for review
- vLLM version: v0.13.0
- vLLM main:
d68209402d

---------

Signed-off-by: 房建伟 <fangjianwei@fangjianweideMacBook-Air.local>
Signed-off-by: fangjianwei <f30058701@china.huawei.com>
Signed-off-by: DreamerLeader <88812830+DreamerLeader@users.noreply.github.com>
Co-authored-by: 房建伟 <fangjianwei@fangjianweideMacBook-Air.local>
Co-authored-by: fangjianwei <f30058701@china.huawei.com>

docs/source/user_guide/feature_guide/kv_pool.md

@@ -305,7 +305,7 @@ Long question:
 curl -s http://localhost:8000/v1/completions -H "Content-Type: application/json" -d '{ "model": "/xxxxx/Qwen2.5-7B-Instruct", "prompt": "Given the accelerating impacts of climate change—including rising sea levels, increasing frequency of extreme weather events, loss of biodiversity, and adverse effects on agriculture and human health—there is an urgent need for a robust, globally coordinated response. However, international efforts are complicated by a range of factors: economic disparities between high-income and low-income countries, differing levels of industrialization, varying access to clean energy technologies, and divergent political systems that influence climate policy implementation. In this context, how can global agreements like the Paris Accord be redesigned or strengthened to not only encourage but effectively enforce emission reduction targets? Furthermore, what mechanisms can be introduced to promote fair and transparent technology transfer, provide adequate financial support for climate adaptation in vulnerable regions, and hold nations accountable without exacerbating existing geopolitical tensions or disproportionately burdening those with historically lower emissions?", "max_completion_tokens": 256, "temperature":0.0 }'
 ```
 
-### Colocation Scenario
+### PD-Mixed Inference
 
 #### 1.Run Mixed Department Script
 
@@ -363,3 +363,9 @@ Long question:
 ```shell
 curl -s http://localhost:8100/v1/completions -H "Content-Type: application/json" -d '{ "model": "/xxxxx/Qwen2.5-7B-Instruct", "prompt": "Given the accelerating impacts of climate change—including rising sea levels, increasing frequency of extreme weather events, loss of biodiversity, and adverse effects on agriculture and human health—there is an urgent need for a robust, globally coordinated response. However, international efforts are complicated by a range of factors: economic disparities between high-income and low-income countries, differing levels of industrialization, varying access to clean energy technologies, and divergent political systems that influence climate policy implementation. In this context, how can global agreements like the Paris Accord be redesigned or strengthened to not only encourage but effectively enforce emission reduction targets? Furthermore, what mechanisms can be introduced to promote fair and transparent technology transfer, provide adequate financial support for climate adaptation in vulnerable regions, and hold nations accountable without exacerbating existing geopolitical tensions or disproportionately burdening those with historically lower emissions?", "max_completion_tokens": 256, "temperature":0.0 }'
 ```
+
+Note: For MooncakeStore, it is recommended to perform a warm-up phase before running actual performance benchmarks.
+
+This is because HCCL one-sided communication connections are created lazily after the instance is launched when Device-to-Device communication is involved. Currently, full-mesh connections between all devices are required. Establishing these connections introduces a one-time time overhead and persistent device memory consumption (4 MB of device memory per connection).
+
+**For warm-up, it is recommended to issue requests with an input sequence length of 8K and an output sequence length of 1, with the total number of requests being 2–3× the number of devices (cards/dies).**

vllm_ascend/distributed/kv_transfer/kv_pool/ascend_store/pool_scheduler.py

@@ -159,6 +159,7 @@ class KVPoolScheduler:
             self._request_trackers.pop(finished_req_id, None)
             self._unfinished_requests.pop(finished_req_id, None)
             self._unfinished_request_ids.discard(finished_req_id)
+            self._preempted_req_ids.discard(finished_req_id)
 
         for req_id in scheduler_output.preempted_req_ids:
             self._preempted_req_ids.update(scheduler_output.preempted_req_ids)