[LoopVectorize] Fails to generate VNNI vpdpbusd for u8xi8 dot product loops

for simple u8xi8 dot product loops, gcc can generate expected vpdpbusd instructions while clang can only generate vpdpwssd.
```c++
int32_t dot_u8_i8(const uint8_t *a, const int8_t *b, int n) {
int32_t sum = 0;
for (int i = 0; i < n; i++) {
sum += (int32_t)a[i] * (int32_t)b[i];
}
return sum;
}
```
gcc: https://godbolt.org/z/4d678xfn8
clang: https://godbolt.org/z/G8sr99PMc