[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/quantization/w8a8.py

@@ -21,6 +21,7 @@ import torch
 import torch_npu
 from vllm.attention.backends.abstract import AttentionType
 from vllm.distributed.parallel_state import get_ep_group
+from vllm.forward_context import get_forward_context
 
 from vllm_ascend.attention.attention_v1 import AscendAttentionState
 from vllm_ascend.ops.moe.experts_selector import select_experts
@@ -97,11 +98,32 @@ class AscendW8A8LinearMethod:
         tp_rank: Optional[int] = 0,
     ) -> torch.Tensor:
         if x.dtype != torch.int8:
+            attn_weight_map = {
+                "AscendQKVParallelLinear": "qkv",
+                "AscendRowParallelLinear": "o",
+            }
+            layer_cls_name = layer.__class__.__name__
+            weight_prefetch_method = get_forward_context(
+            ).weight_prefetch_method
+            assert weight_prefetch_method is not None
+
+            # prefetch_qkvo_proj.weight preprocess
+            weight_prefetch_method.maybe_prefetch_attn_weight_preprocess(
+                prefix=attn_weight_map.get(layer_cls_name, ""),
+                weight=layer.weight,
+                start_flag=x,
+            )
+            # quant
             x = quant_per_tensor(
                 x,
                 layer.aclnn_input_scale_reciprocal,
                 layer.aclnn_input_offset,
             )
+            # prefetch_qkvo_proj.weight postprocess
+            if layer_cls_name in attn_weight_map.keys():
+                weight_prefetch_method.maybe_prefetch_attn_weight_postprocess(
+                    x)
+
         quant_bias = layer.quant_bias if tp_rank == 0 else None
         if is_310p():
             # On 300I Duo platform, we need transpose again if