[BUG] i64数据经pnnx存档后，无法在forward里正确输出

[Baiyuetribe]

复现步骤

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
    def forward(self, x):
        i32 = torch.tensor([2, 1, 3, 4], dtype=torch.int32)
        i64 = torch.tensor([2, 1, 3, 4], dtype=torch.int64)
        return i32, i64

pnnx demo.pt inputshape=[32]
经过上述转换，ncnn.param内容为

7767517
2 2
MemoryData               pnnx_fold_11             0 1 out0 0=4
MemoryData               pnnx_fold_22             0 1 out1 0=4

ncnn.bin文件内容为

最终，在cpp中提取时，i64数据输出错误：
int|int32_t ---> [2,0,1,0]
int64_t ---> [2,1,2,0]

故障来源

torch.gather实现上，Python强制要求i64类型，然后经pnnx后，ncnn无法正确输出int结果

[Baiyuetribe]

#include "net.h"

int main()
{
    auto net = ncnn::Net();
    net.load_param("demo.ncnn.param");
    net.load_model("demo.ncnn.bin");
    ncnn::Extractor ex = net.create_extractor();

    ncnn::Mat out0, out1;
    ex.extract("out0", out0);
    ex.extract("out1", out1);
    int* p = out0;
    printf("out0 = %d,%d,%d,%d\n", p[0], p[1], p[2], p[3]);
    int* p1 = out1;
    printf("out1 = %d,%d,%d,%d\n", p1[0], p1[1], p1[2], p1[3]);
    return 0;
}

输出：

out0 = 2,1,3,4 # i32正确输出
out1 = 2,0,1,0 # i64 错误输出

结论，pnnx存档时，i64应该强制转i32存档。如果不作为Memorydata存档，那么输入i64也要做一次i32转换

[Baiyuetribe]

fix: #5881