[1/N][Feat] Add weight prefetch feature for Attention layers (#3146)

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

- Refacotr and integrate a unified `WeightPrefetchMethod`
- Integrate `qkv_proj.weight` and `o_proj.weight` in quantized Attention
modules
- Prefetching these weights ahead of matmul-like operators imporves
performance by reducing L2 cache transfer latency

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

Add a new config in `--additional-config` for configuration:
```json
{
    "weight_prefetch_config": {
        "enabled": false,
        "prefetch_ratio": {
            "attn": {
                "qkv": 1.0,
                "o": 1.0,
            },
        },
    },
}
```
This feature is enabled by default, and can be disabled through this
configuration

### How was this patch tested?


- vLLM version: v0.11.0

---------

Signed-off-by: yuzhup <15705211260@163.com>
Signed-off-by: zhoux77899 <zhouxiang100@huawei.com>
Co-authored-by: yuzhup <15705211260@163.com>

vllm_ascend/ascend_config.py

@@ -45,6 +45,12 @@ class AscendConfig:
             "ascend_scheduler_config", {})
         self.ascend_scheduler_config = AscendSchedulerConfig(
             ascend_scheduler_config)
+
+        weight_prefetch_config = additional_config.get(
+            "weight_prefetch_config", {})
+        self.weight_prefetch_config = WeightPrefetchConfig(
+            weight_prefetch_config)
+
         # Todo: Once https://github.com/vllm-project/vllm/issues/22246 is merged in vllm. Remove this config
         self.expert_map_path = additional_config.get("expert_map_path", None)
         self.expert_map_record_path = additional_config.get(
@@ -65,7 +71,6 @@ class AscendConfig:
         ) and not self.torchair_graph_config.enabled and vllm_config.parallel_config.enable_expert_parallel
         self.multistream_overlap_shared_expert = additional_config.get(
             "multistream_overlap_shared_expert", False)
-        self.enable_prefetch = additional_config.get("enable_prefetch", False)
         self.lmhead_tensor_parallel_size = additional_config.get(
             "lmhead_tensor_parallel_size", None)
         if self.lmhead_tensor_parallel_size is not None:
@@ -185,6 +190,24 @@ class AscendSchedulerConfig:
                 setattr(self, k, v)
 
 
+class WeightPrefetchConfig:
+    """
+    Configuration Object for weight_prefetch_config from additional_config
+    """
+
+    prefetch_ratio: dict = {
+        "attn": {
+            "qkv": 1.0,
+            "o": 1.0,
+        },
+    }
+
+    def __init__(self, weight_prefetch_config: dict):
+        self.enabled = weight_prefetch_config.get("enabled", False)
+        self.prefetch_ratio = weight_prefetch_config.get(
+            "prefetch_ratio", self.prefetch_ratio)
+
+
 _ASCEND_CONFIG: Optional[AscendConfig] = None