Improved performance when multi-threading

This release introduces a performance enhancement by integrating a persistent thread pool to manage multi-threading. Previously, the algorithm created and destroyed threads in each boosting step, which incurred overhead. The performance gain is largest on smaller datasets.