add clang-7 benchmark numbers

esp. memcpy_s is significantly faster with
clang-7 on darwin. Only 50-55% slowdown, not 90%.
And with clang-4+ -march=native the overhead is only 5-20%.