[llvm] Revert "[AMDGPU] Elide bitcast fold i64 imm to build_vector" (PR #160325)

[Janek van Oirschot via llvm-commits]

JanekvO wrote:

Resulted in error when compiling the following for gfx942:
```
define amdgpu_kernel void @foo(ptr %0) {
entry:
  %1 = getelementptr i8, ptr %0, i64 4
  store i32 0, ptr %0, align 4
  store i64 0, ptr %1, align 4
  ret void
}
```

https://github.com/llvm/llvm-project/pull/160325