[VectorCombine] failed to fold fptoui(fdiv(uitofp(x),uitofp(y))) into udiv(x,y) when target has native vector udiv

https://godbolt.org/z/eafrhETE3
ref: https://github.com/llvm/llvm-project/issues/205305#issuecomment-4787165610
```c
typedef unsigned int u __attribute__((__vector_size__(16)));
typedef double d __attribute__((__vector_size__(32)));

u a(u x, u y){
d a = __builtin_convertvector(x, d);
d b = __builtin_convertvector(y, d);
d r = a / b;
return __builtin_convertvector(r, u);
}

u b(u x, u y){
return x / y;
}
```
```asm
a:
ushll v2.2d, v0.2s, #0
ushll v3.2d, v1.2s, #0
ushll2 v0.2d, v0.4s, #0
ushll2 v1.2d, v1.4s, #0
ucvtf v2.2d, v2.2d
ucvtf v3.2d, v3.2d
ucvtf v0.2d, v0.2d
ucvtf v1.2d, v1.2d
fdiv v2.2d, v2.2d, v3.2d
fdiv v0.2d, v0.2d, v1.2d
fcvtzu v1.2d, v2.2d
fcvtzu v0.2d, v0.2d
uzp1 v0.4s, v1.4s, v0.4s
ret

b:
ptrue p0.s, vl4
udiv z0.s, p0/m, z0.s, z1.s
ret
```

This is particularly important for vector cases, as u32->f64->u32 will halve the throughput.