`memcmp` -> `bcmp` optimization not firing when fed to bitwise or

Testcase:
```c++
char *p;
bool b;
int main() {
b |= __builtin_memcmp(p, "1234567", 7);
}
```

This produces unnecessarily slow code, which computes not only whether the strings are equal but also whether they are greater or less: https://godbolt.org/z/Tf71qovK3

It looks like the transform from the `memcmp` to `bcmp` lib function is not firing here, though it should be able to.

Start with the provided C++ testcase using __builtin_memcmp in a bitwise-or expression, and reproduce the generated code from the linked Godbolt example. Trace the memcmp-to-bcmp optimization and verify that the completed change makes this case use the bcmp form without computing ordering information.

Written by the indexing model from the issue text.