[libc++] Use __libcpp_clz for a tighter __log2i function

While looking at D113413 I noticed that __log2i could perhaps be
improved to be both slightly smaller and faster.

projects/libcxx/benchmarks/sort.libcxx.out --benchmark_filter=BM_Sort_uint32_QuickSortAdversary*
suggests this is performance neutral, but it shaves a few bytes off the
benchmark binary, and even more off a Chromium build.

Differential revision: https://reviews.llvm.org/D125958