[2/N][Feat] Attention and MoE weight prefetch in Qwen3MoE models (#3203)

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

- Refacotr and integrate a unified `WeightPrefetchMethod`
- Integrate `gate_up_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": True,
        "prefetch_ratio": {
            "moe": {
                "gate_up": 0.8
            },
        },
    },
}
```
This feature is enabled by default, and can be disabled through this
configuration

### How was this patch tested?


- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

---------

Signed-off-by: yuzhup <15705211260@163.com>

tests/e2e/singlecard/ops/test_fused_moe.py

@@ -291,7 +291,9 @@ def test_select_experts(
         custom_routing_function.return_value = (mock_weights, mock_ids)
 
     with patch("vllm_ascend.ops.moe.experts_selector._native_grouped_topk"
-               ) as mock_native_grouped_topk:
+               ) as mock_native_grouped_topk, \
+            patch('vllm_ascend.ops.moe.experts_selector.get_forward_context',
+                  return_value=MagicMock(weight_prefetch_method=MagicMock())):
         mock_native_grouped_topk.side_effect = lambda x, num_groups, k: torch.randn_like(
             x)
 
@@ -325,7 +327,9 @@ def test_select_experts(
 
 @pytest.mark.parametrize("device", DEVICE)
 def test_select_experts_invalid_scoring_func(device: str):
-    with pytest.raises(ValueError,
+    with patch('vllm_ascend.ops.moe.experts_selector.get_forward_context',
+                  return_value=MagicMock(weight_prefetch_method=MagicMock())), \
+            pytest.raises(ValueError,
                        match="Unsupported scoring function: invalid"):
         select_experts(hidden_states=torch.randn(1, 128, device=device),
                        router_logits=torch.randn(1, 8, device=device),