[libc++] Optimize nth_element for small ranges

`std::nth_element` can be implemented very efficiently when we know at compile-time that we have a range of 3 or 5 elements. We could switch on whether we know the size of the range at compile-time, and if so, use a specialized algorithm for that.

Start by locating the libc++ implementation and tests for std::nth_element, then inspect how compile-time range sizes are represented. The work is complete when ranges of 3 or 5 elements use specialized algorithms without changing nth_element behavior, with performance and correctness coverage added.

Written by the indexing model from the issue text.