Use Rayon for escore algorithm guessing.

[ZakisM]

Hey, so I tried using Rayon inside the escore algorithm guess method.

I was able to get quite a nice speedup on my 4 core PC:

Benchmark 1: roget_old.exe
  Time (mean ± σ):      3.778 s ±  0.036 s    [User: 0.005 s, System: 0.004 s]
  Range (min … max):    3.708 s …  3.821 s    10 runs

Benchmark 2: roget_new.exe
  Time (mean ± σ):     650.0 ms ±  11.2 ms    [User: 1.3 ms, System: 6.8 ms]
  Range (min … max):   633.3 ms … 662.5 ms    10 runs

Summary
  'roget_new.exe' ran
    5.81 ± 0.11 times faster than 'roget_old.exe'

However, I noticed that each time I run the program with these changes the score is slightly different, i'm assuming that i'm doing something incorrect with Atomics?

[jonhoo]

Hi! Thanks for contributing :)

I think I'd approach this slightly differently — rather than having the threads all communicate using atomics, instead have each thread operate completely independently and compute their own "best" candidate, and then do a quick (single-threaded) pass at the end that just finds the best among those found by each thread. I think you should be able to do that just with rayon (probably using something like ParallelIterator::reduce).

Also, I just landed some pretty big restructuring changes on main, so you may want to do your changes on top of that instead :)