[libc++] Consider implementing a new hashing algorithm

We currently use either murmur2 or CityHash, depending on `sizeof(size_t)`. We should look into providing a new, faster, hashing algorithm.

The issue names libc++ and its current murmur2 and CityHash choices, but no files, tests, algorithm, benchmark, or acceptance criteria. Start by locating the libc++ hashing implementation and reviewing how the size_t-dependent selection works; done requires a defined faster algorithm and evidence that it is suitable for both configurations.

Written by the indexing model from the issue text.