Branchless scalar array-array union

[saik0]

https://lemire.me/blog/2021/07/14/faster-sorted-array-unions-by-reducing-branches/

[saik0]

impl at: saik0@f0b3f02

Perf delta range +12% -25% in runtime (lower is better). I'm going to wait until after #12 to re-evaluate before opening a PR.

I suspect the branch predictor is doing us some favors for runs, which will be less likely in run containers, then this will be a "difference that makes a difference"