Investigate `llvm::sys::getHostCPUName()` and `llvm::sys::getHostCPUFeatures()` to possibly improve code generation

[erick-xanadu]

We currently set these to "generic" and "" respectively. While we have the targetMachine->setOptLevel(llvm::CodeGenOptLevel::None); to avoid large compilation times, I believe that using the host cpu name and host cpu features may provide some performance benefits.

[mlxd]

This is a nice idea. I'd expect enabling all features to potentially cause some regressions, so choosing the optimal set may require some combination exploring.

I've added a quick 'n' dirty capture utility around the above for CLI querying the supported CPU set options. Maybe identifying the right ones to enable could start with the coarse-grained (ie more general) instructions (e.g. avx/avx2 on X86), and later moving to more targeted ones (e.g. avx512vbmi2 on X86).

https://gist.github.com/mlxd/eb4ee6af14f50476db3bc7088631ee40