Enable LTO

[springmeyer]

For projects that use node-cpp-skel and end up having lots of .cpp files, LTO starts to become potentially important for best performance per https://github.com/mapbox/cpp/blob/master/glossary.md#link-time-optimization.

This ticket tracks getting an LTO example working in node-cpp-skel that other modules can follow. This is worthwhile but a little hairy since it requires:

• the same compiler to be used for compiling all binaries (tricky with mason unless you ask mason to source compile)
• it requires upgrading binutils from mason to use the "gold" linker within binutils for linux

Ideally as part of this effort we also document an exact, measurable situation when LTO actually helps. Having it help performance in theory is not worth the setup overhead, so we'll want to have a clear demonstration it helps perf. Although one caveat is that we should likely investigate using -fsanitize=cfi in production in the future and that requires LTO: (https://blog.trailofbits.com/2016/10/17/lets-talk-about-cfi-clang-edition/ and https://clang.llvm.org/docs/ControlFlowIntegrity.html).

Related:

• vtquery port (fairly simply since vtquery has no binary dependencies): https://github.com/mapbox/vtquery/compare/lto
• carmen-cache issue : Refactor followup: enable LTO carmen-cache#117
• carmen-cache followup fixes: Fix LTO on linux carmen-cache#119

[springmeyer]

now enabled after 93edf4f

[springmeyer]

Ideally as part of this effort we also document an exact, measurable situation when LTO actually helps.

Never did this. Still worthwhile. The reason I merged without doing a perf validation benchmark is that:

1. I noticed that adding -flto can catch ODR errors that are otherwise silent. So making sure we can compile with LTO increases the robustness of the code
2. I wanted to support fsanitize=cfi which requires it