Test BenchmarkDotNet against unstable/multimodal benchmarks from CoreCLR/CoreFX repo

[adamsitnik]

We are considering switching from xunit-performance to BenchmarkDotNet for CoreCLR and CoreFx repositories.

One of the requirements is to test BDN against the most problematic benchmarks:
https://github.com/dotnet/coreclr/issues/11657#issuecomment-365743836
https://github.com/dotnet/coreclr/issues/16613#issuecomment-371589608

• port the benchmarks to BenchmarkDotNet
• run them every 6 hours for at least 2 weeks for same .NET Core version
• store the results
• analyze the results
• share the results + analyzis

/cc @AndyAyersMS

[adamsitnik]

Done, everything available in a new repo: https://github.com/dotnet/performance/

+-20 of the 2000 benchmarks are not stable. Mostly multithreaded benchmarks (which rely on ThreadPool heuristics), loop body alignment (for benchmarks that do have their own loop), array body alginement