nodebug for HWY_API, attribute debug information to call site

I was examining some code that uses Highway and noticed that in perf annotate, the instructions from Highway functions are all attributed to Highway source files (unsurprisingly). I find this a little suboptimal, as it's not always obvious which instruction goes with which caller line, and this is a bit different than direct intrinsic use.

There's another option: if inline functions are marked with the nodebug attribute, the Highway function line information is instead attributed to the caller. Maybe we should be able to use the [artificial attribute](https://groups.google.com/g/llvm-dev/c/cgGxEmxALqo), but that is apparently not correctly implemented.

If there are no concerns or objections in the next couple of weeks, I'm planning on marking HWY_API with nodebug under gcc/clang to make the debug information attribution more similar to direct intrinsic use.