[llvm] [LV] Improve AnyOf reduction codegen. (PR #78304)

[Arthur Eubanks via llvm-commits]

aeubanks wrote:

it looks like the C++ code getting miscompiled is
```
  uint32_t character_bytes = 0;
  bool all_same_length = true;
  for (const auto& value : input_data) {
    if (value.size() != input_data.front().size()) {
      all_same_length = false;
    }
    character_bytes += value.size();
  }
```

`all_same_length` is `false` with `-fno-vectorize` but `true` when vectorizing

https://github.com/llvm/llvm-project/pull/78304