[llvm] Emit float128 libcalls for some F128 variants (PR #79611)

[Fangrui Song via llvm-commits]

MaskRay wrote:

This LGTM, but we'd need `__builtin_*f128` clang/test/CodeGen tests.

My digging into the log says that https://github.com/llvm/llvm-project/commit/4bf47bc9d7cad0fbec78df8816d9b4ae7ae2be30 (2013) introduced these l libcall names. aarch64 long double is quad precision and its `l` suffix (for long double) does indicate quad precision. 

https://github.com/llvm/llvm-project/pull/79611