Function init_cpu_pool 

pub fn init_cpu_pool() -> Option<usize>

Builds the global rayon pool with an explicit width and an explicit QoS, so neither depends on which thread first touched rayon. Safe to call more than once and from either binary; a pool that already exists is left alone.

Returns the thread count the pool was built with, or None if the global pool already existed. Builds the dedicated GEMV pool (and, for legacy callers that still touch rayon::prelude on the global pool, a matching global pool). Prefer for_each_row / for_each_chunk_init so matvecs stay on the dedicated pool — mixing both pools oversubscribes P-cores.

Returns the thread count the global pool was built with, or None if the global pool already existed.