af::max returns the right maximum value but the wrong idx when the input is a subarray.

Using copy() on the subarray fixes the problem.

It happens with CUDA, OpenCL and CPU backends,

Description

• Additional details regarding the bug
• Did you build ArrayFire yourself or did you use the official installers
  Official installer
• Which backend is experiencing this issue? (CPU, CUDA, OpenCL)
  CPU, CUDA, OpenCL
• Do you have a workaround?
  Copy the subarray
• Can the bug be reproduced reliably on your system?
  yes

-->

Reproducible Code and/or Steps

`
#include <arrayfire.h>
#include

int main(int argc, char const* argv[]) {

try {

    af::setBackend(AF_BACKEND_CUDA);
    af::setDevice(0);
    af::info();

    float input_vals[25] = {0.0168, 0.0278, 0.0317, 0.0248, 0.0131, 0.0197, 0.0321, 0.0362, 0.0279,
                            0.0141, 0.0218, 0.0353, 0.0394, 0.0297, 0.0143, 0.0224, 0.0363, 0.0404,
                            0.0302, 0.0142, 0.0217, 0.0355, 0.0398, 0.0302, 0.0144};
    af::array input(5, 5, input_vals);

    af::print("input", input);

    unsigned idx;
    float max_val;
    af::max<float>(&max_val, &idx, input);

    std::cout << "idx: " << idx << ", max_val: " << max_val << std::endl; //OK

    unsigned row = idx % input.dims(0);
    unsigned col = idx / input.dims(0);

    af::array input_cropped = input(af::seq(row - 1, row + 1), af::seq(col - 1, col + 1));

    af::print("input_cropped", input_cropped);

    af::max<float>(&max_val, &idx, input_cropped);

    std::cout << "idx: " << idx << ", max_val: " << max_val << std::endl; //NOK, wrong idx (6 instead of 4), right maximum value

    af::print("input_cropped.copy()", input_cropped.copy());

    af::max<float>(&max_val, &idx, input_cropped.copy());

    std::cout << "idx: " << idx << ", max_val: " << max_val << std::endl; //OK

} catch (const af::exception& e) {
    std::cerr << e.what() << '\n';
}

return 0;

}
`

.\test.exe
ArrayFire v3.9.0 (CUDA, 64-bit Windows, build b59a1ae)
Platform: CUDA Runtime 12.2, Driver: 12080
[0] NVIDIA RTX A2000 8GB Laptop GPU, 8192 MB, CUDA Compute 8.6
input
[5 5 1 1]
0.0168 0.0197 0.0218 0.0224 0.0217
0.0278 0.0321 0.0353 0.0363 0.0355
0.0317 0.0362 0.0394 0.0404 0.0398
0.0248 0.0279 0.0297 0.0302 0.0302
0.0131 0.0141 0.0143 0.0142 0.0144

idx: 17, max_val: 0.0404

input_cropped
[3 3 1 1]
0.0353 0.0363 0.0355
0.0394 0.0404 0.0398
0.0297 0.0302 0.0302

idx: 6, max_val: 0.0404

input_cropped.copy()
[3 3 1 1]
0.0353 0.0363 0.0355
0.0394 0.0404 0.0398
0.0297 0.0302 0.0302

idx: 4, max_val: 0.0404

System Information

1. ArrayFire version
   3.9.0 on Windows 10

Checklist

• [ x ] Using the latest available ArrayFire release
• [ x ] GPU drivers are up to date