pub fn optimal_concurrency() -> usize
Optimal core count for compute work: P-cores on Apple Silicon, physical on x86.