Improve performance by reverting Apple silicon workaround (root problem is fixed upstream)

[jackryanservia]

Closes: #1337 ;)
Bindings PR: o1-labs/o1js-bindings#251

• Spawns one worker for each physical core by default (for some reason, numCpus - 1 was faster a month ago, but when I tested today, numCores was slightly faster).
• Exposes setNumberOfWorkers() to allow developers to manually set the number of workers.
• Removes detect-gpu dependency

I tested and everything worked well, prover time decrease of ~35% in browser, and ~46% in node. Compile time improvement is even more significant ~60%! (Tested on Apple M1 Pro 10-Core).

I wasn't exactly sure of the best way to implement the setNumberOfWorkers() function, and where everything should live/what it should be named. So I would appreciate any feedback on that kinda thing :)

[mitschabaude]

great to have this in!

[dfstio]

My tests show that on M2 Max which has 12 cores, the optimal number of workers is 8, it is 35% faster than with 12 workers.
System info shows:
Total Number of Cores: 12 (8 performance and 4 efficiency)

Can you please set the default number of workers for M2 Max to 8? It seems like 12 now.

[mitschabaude]

Can you please set the default number of workers for M2 Max to 8? It seems like 12 now.

I don't think this is easily possible because the API that nodejs gives us is availableParallelism() which returns a number of cores. It doesn't distinguish between performance and efficiency cores.

I suggest to just set the number of workers yourself for fine-tuning on particular architectures, seems a bit much for us to test and optimize every possible hardware