[clang] [Sema] Mark alias/ifunc targets used and consider mangled names (PR #87130)

[Fangrui Song via cfe-commits]

================
@@ -0,0 +1,47 @@
+// RUN: %clang_cc1 -triple %itanium_abi_triple -Wunused -x c -verify %s
+// RUN: %clang_cc1 -triple %itanium_abi_triple -Wunused -x c++ -verify=expected,cxx %s
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+static int f(void) { return 42; }
+int g(void) __attribute__((alias("f")));
----------------
MaskRay wrote:

> Ah! right you pointed this out in your note. Perhaps we should fix/change that first in clang?

I remember an existing issue about the mangling difference, but I don't recall the link now.
I have analyzed it a bit and run into a feature where we need mangling.
The discrepancy from GCC is indeed unfortunate and in practice makes `alias`/`ifunc` not useful for internal linkage targets in C++.

https://eel.is/c++draft/dcl.link gives an example

```cpp
extern "C" {
  static void f4();             // the name of the function f4 has internal linkage,
                                // so f4 has no language linkage; its type has C language linkage
}
```

I am unsure whether "no language linkage" means that ignoring the C language linkage is fine.

---

Improving GCC consistency is not the goal of this PR. This PR intends to make clangSema and clangCodeGen consistent. If we want to follow GCC in the future, the updated test will show the differences.

https://github.com/llvm/llvm-project/pull/87130