Performance optimizations for `co.py`

[cliffckerr]

• Add parallelization (can use sc.parallelize()) at the appropriate place -- over counties, I guess? Or maybe in optimize() over for mask, targ_row, n in zip(samp_masks, range(len(targs)), n_hhs)?
• Could try to add Numba to FTdist() and anneal() if they're a computational bottleneck -- that could recover Julia-speed performance if this version is significantly slower (but I don't recall that it's that much slower).
• Could also just point Claude etc at this file and see what it can find :)