clang builds not using full LTO

[Fidget-Spinner]

GCC defaults to full LTO with --with-lto=yes passed. Clang defaults to ThinLTO with --with-lto=yes passed. You need --with-lto=full to enable full LTO.

[Fidget-Spinner]

This might explain the slowdown on macOS and the rest. It has been observed in the past to cause a tremendous slowdown for macOS python/cpython#122580.

This also means the tail-calling interpreter without with ThinLTO is already significantly faster than LTO-full 👀 .