AArch64 SME SMOPA (4-way) outer product instruction gives incorrect result
Description of problem:
The SME SMOPA (4-way) instruction ([spec](https://developer.arm.com/documentation/ddi0602/2023-09/SME-Instructions/SMOPA--4-way---Signed-integer-sum-of-outer-products-and-accumulate-?lang=en)) is giving incorrect result. Example below for 8-bit variant, which is equivalent to following Python example (128-bit VL) to make it clearer:

```
import numpy as np
vl = 128
esize = 32
dim = vl // esize

A = range(16)
B = range(16, 32)
C = np.zeros((4, 4,), dtype=np.int32)

for row in range(dim):
    for col in range(dim):
        for k in range(4):
            C[row, col] += A[4*row + k] * B[4*col + k]

print(C)

[[ 110  134  158  182]
 [ 390  478  566  654]
 [ 670  822  974 1126]
 [ 950 1166 1382 1598]]
```

main.c
```
#include <stdio.h>
#include <stdint.h>

void foo(int *dst);

int main() {
  int32_t dst[16];
  foo(dst);

  // This should print:
  // >>> 110  134  158  182
  // >>> 390  478  566  654
  // >>> 670  822  974  1126
  // >>> 950  1166  1382  1598
  for (int i=0; i<4; ++i) {
    printf(">>> ");
    for (int j=0; j<4; ++j) {
      printf("%d  ", dst[i * 4 + j]);
    }
    printf("\n");
  }
}
```

foo.S

```
.global foo
foo:
  stp x29, x30, [sp, -80]!
  mov x29, sp
  stp d8, d9, [sp, 16]
  stp d10, d11, [sp, 32]
  stp d12, d13, [sp, 48]
  stp d14, d15, [sp, 64]

  smstart

  ptrue p0.b
  index z0.b, #0, #1
  mov   z1.d, z0.d
  add   z1.b, z1.b, #16

  zero  {za}
  smopa za0.s, p0/m, p0/m, z0.b, z1.b

  // Read the first 4x4 sub-matrix of elements from tile 0:
  mov w12, #0
  mova z0.s, p0/m, za0h.s[w12, #0]
  mova z1.s, p0/m, za0h.s[w12, #1]
  mova z2.s, p0/m, za0h.s[w12, #2]
  mova z3.s, p0/m, za0h.s[w12, #3]

  // And store them to the input pointer (dst in the C code):
  st1w {z0.s}, p0, [x0]
  add x0, x0, #16
  st1w {z1.s}, p0, [x0]
  add x0, x0, #16
  st1w {z2.s}, p0, [x0]
  add x0, x0, #16
  st1w {z3.s}, p0, [x0]

  smstop

  ldp d8, d9, [sp, 16]
  ldp d10, d11, [sp, 32]
  ldp d12, d13, [sp, 48]
  ldp d14, d15, [sp, 64]
  ldp x29, x30, [sp], 80
  ret
```
Steps to reproduce:
```
$ clang -target aarch64-linux-gnu -march=armv9-a+sme main.c foo.S
$ ~/qemu/build/qemu-aarch64 -cpu max,sme128=on a.out
>>> 110  478  158  654
>>> 0  0  0  0
>>> 670  1166  974  1598
>>> 0  0  0  0
```
Additional information: