Use target_clones for count_ones, rather than duplicating code.

Don't clone crc64 overloads that we don't need to (that are just calling
another function).  Otherwise, we can run into trouble when tdaq code
loads libraries with RTLD_NOW.  (That's because with RTLD_NOW, the ifunc
resolver may be called during reloc processing.  If we get an instance
of the resolver in a different library than being resolved, then
we may end up looking at a GOT entry that hasn't yet been filled in.)


Former-commit-id: 62cb2ef6