How does Mojo's parallelize function distribute sub-tasks across CPU cores?

[JackKelly]

Does parallelize spin up a pool of worker threads, one thread per CPU core, and use work stealing to balance the tasks across threads? (like Rayon?)

Or does parallelize spin up one OS thread per task? Or use a blocking queue to distribute tasks to a thread pool?

[Brian-M-J]

I would also really like to know about a lot of details of Mojo's internals. I've heard the team mention their ambitions about the Actor model. Maybe each task is implemented as an actor internally.