[X86] no longer producing vector shifts after instcombine change

```c
#include

__m128i s32(__m128i a){
return _mm_srai_epi32(_mm_slli_epi32(a, 31),31);
}
__m128i s16(__m128i a){
return _mm_srai_epi16(_mm_slli_epi16(a, 15),15);
}
```

regressed since e71870512fd896bf6cf34e8ae650f4cf20923258

https://godbolt.org/z/W88oGfExK

Start with the C reproducer and compare its output on Godbolt before and after commit e71870512fd896bf6cf34e8ae650f4cf20923258. Trace the instcombine change affecting the two SIMD shift sequences and verify that compilation produces vector shifts again for both s32 and s16.

Written by the indexing model from the issue text.