Implement multithreaded push_many on wt_string

[chrisbarber]

Implementation of the concept in #13, along with an optimization I came across in packed_vector.

[chrisbarber]

I haven't had time to dig further but the optimization in packed_vector gave a 5-10x speedup when wavelet tree construction ends up being all inserts at the ends of bitvectors. I think shift_right is the culprit, which should effectively be a noop when inserting at the end (pushing back).

I left the threading feature OFF by default, so the normal build steps should still work.