AddRmsNorm + quant 支持pattern算子融合

背景

当前已有 npu_add_rms_norm，npu_quantize以及npu_add_rms_norm_quant算子
且torchair目前具备如下能力：
对于aclgraph，想实现特定算子的融合，可以借助pytorch原生的pattern_matcher逻辑，用融合算子替换原来的多个算子（实际上是fx图多个节点的替换）

本合入则是利用该能力注册实现对npu_add_rms_norm+quantize 计算进行融合

融合前

graph LR
    subgraph IN[" "]
        direction LR
        X1["x1"]
        X2["x2"]
        G["gamma"]
    end
    
    subgraph OP[" "]
        direction LR
        A["npu_add_rms_norm"]
        B["npu_quantize"]
    end
    
    subgraph OUT[" "]
        direction LR
        Y1["y_out"]
        Y2["x_out"]
    end
    
    X1 --> A
    X2 --> A
    G --> A
    A --> B
    A --> Y2
    B --> Y1
    
    style A fill:#e3f2fd,stroke:#1976d2
    style B fill:#e3f2fd,stroke:#1976d2

融合后

    graph LR
    subgraph IN[" "]
        direction LR
        X1["x1"]
        X2["x2"]
        G["gamma"]
        S["scales"]
        Z["zero_points"]
    end
    
    subgraph OP[" "]
        FUSED["npu_add_rms_norm_quant"]
    end
    
    subgraph OUT[" "]
        direction LR
        Y1["y1"]
        Y2["x_out"]
    end
    
    X1 --> FUSED
    X2 --> FUSED
    G --> FUSED
    S --> FUSED
    Z --> FUSED
    FUSED --> Y1
    FUSED --> Y2
    style FUSED fill:#c8e6c9,stroke:#388e3c