NCNN RVV int8 and fp16sa optimization

[Xinyu302]

复现环境

• 官方buildroot镜像，55MB 内存+256MB swap内存（不开启swap时部分模型的在55MB内存下运行比较极限，效果不如开启swap内存）。
• 编译ncnn：

$ cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/c906-v240-musl.toolchain.cmake -DCMAKE_BUILD_TYPE=release -DNCNN_BENCHMARK=OFF -DNCNN_OPENMP=OFF -DNCNN_THREADS=OFF -DNCNN_RUNTIME_CPU=OFF -DNCNN_RVV=ON -DNCNN_SIMPLEOCV=ON -DNCNN_BUILD_EXAMPLES=ON -DNCNN_BUILD_TESTS=OFF ..
$ make -j 8

测试情况

优化中加入了对Conv的Winograd优化，但是由于该优化会使得模型weight变大，部分模型效果不好，因此修改了benchncnn源码，控制模型是否使用Winograd优化.

此外，除官方说明的可以在55MB下运行的模型外，本优化可以使squeezenet_ssd_int8模型正常运行。

运行命令：

./benchncnn 4 1 0 -1 0

模型运行时间：

          squeezenet  min =  224.48  max =  226.46  avg =  225.52
     squeezenet_int8  min =  540.41  max =  551.01  avg =  545.77
           mobilenet  min =  336.35  max =  338.45  avg =  337.75
      mobilenet_int8  min =  888.55  max =  889.63  avg =  889.00
        mobilenet_v2  min =  235.79  max =  237.13  avg =  236.29
        mobilenet_v3  min =  208.19  max =  209.76  avg =  208.71
          shufflenet  min =  166.60  max =  167.01  avg =  166.78
       shufflenet_v2  min =  156.89  max =  159.10  avg =  158.19
             mnasnet  min =  256.52  max =  257.56  avg =  257.06
     proxylessnasnet  min =  299.75  max =  300.23  avg =  299.93
     efficientnet_b0  min =  365.72  max =  366.99  avg =  366.16
   efficientnetv2_b0  min =  501.54  max =  504.23  avg =  502.88
        regnety_400m  min =  344.23  max =  346.39  avg =  345.04
           blazeface  min =  119.88  max =  120.41  avg =  120.07
      googlenet_int8  min = 2746.41  max = 2748.56  avg = 2747.50
       resnet18_int8  min = 2782.48  max = 2792.17  avg = 2788.06
 squeezenet_ssd_int8  min = 3491.84  max = 3586.81  avg = 3524.24
       mobilenet_ssd  min =  775.98  max =  785.64  avg =  778.90
  mobilenet_ssd_int8  min = 1947.19  max = 1978.77  avg = 1956.45
  mobilenetv2_yolov3  min =  821.81  max =  826.52  avg =  824.13
           nanodet_m  min =  395.29  max =  396.09  avg =  395.72
    yolo-fastest-1.1  min =  185.70  max =  186.40  avg =  186.13
      yolo-fastestv2  min =  159.52  max =  160.26  avg =  159.77
          FastestDet  min =  165.56  max =  168.29  avg =  166.85

[shiptux]

尊敬的参赛选手，您好。
本次锦标赛您所提交的 PR 初步复测结果如 https://github.com/plctlab/rvspoc/blob/main/Results/Verifications/S2310/README.md 所示。如有任何异议请回复本条评论。如确认无误请回复 “确认无误”，感谢您的配合。

[Xinyu302]

确认无误

[shiptux]

确认无误

感谢您的回复。