[clang] [WIP][Clang] Add __builtin_get_counted_by builtin (PR #102549)

[Bill Wendling via cfe-commits]

bwendling wrote:

> So the idea here is that if the struct in question uses counted_by, you automatically set the count... and if, for whatever reason, the compiler can't find the corresponding field, you just throw away the count? That seems like an terrifying API; it's impossible to predict what it will do with a given bit of code.

I agree and it was mentioned in the [GCC bug thread](https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116016). This builtin will have very limited utility (none?) outside of the Linux kernel and it's `kmalloc()` implementation. I would go so far as to place large warnings about its usage in the documentation, or omit documentation altogether.

> Maybe it makes sense to return null if you find the field, but it doesn't have a counted_by attribute at all. But there should still be diagnostics for all the other failure modes.

I agree, with one exception. We're planning on extending the `counted_by` attribute to support pointers in structures. So we'll have to remove the diagnostic in those situations when that support finally lands.

https://github.com/llvm/llvm-project/pull/102549