Muddled the code too much, and only gave a marginal performance improvement in the grand scheme of things. Other ways to parallelize MCTS will be nicer to implement and could yield better results.
- Breaks backwards compat solver code with last version. Concurrent broke backwards compat because of race conditions with rng, but single is now broken too, despite it being 2x faster (!!!!)
- Literally twice as fast as Rust now in single thread
- Concurrent doesn't work yet, deadlocks somewhere..?
Asriel Camora