Revert "Removes dependency to fast CLZ/bitscanforward. Instead. Now robin_hood map is plain C++ code, without any need for intrinsics!"

This reverts commit f45a309a.

Removes dependency to fast CLZ/bitscanforward. Instead. Now robin_hood map is plain C++ code, without any need for intrinsics!

In my benchmarks g++ and clang++ also produce very fast code, faster than the old implementation.

* Add try_emplace from C++17

* Add insert_or_assign from C++17. Fixes #71

This fixes the intrinsics usage on ARM64. Also adds Windows 10 ARM64 compatibility (see: https://docs.microsoft.com/en-us/cpp/intrinsics/arm64-intrinsics?view=vs-2019 )