reverse order of tbb task_scheduler_init and global_control

according to the Intel tbb experts, one should do the tbb::global_control before the tbb::task_scheduler_init

Also improved the logic concerning the size of the thread pool:

• -100: no thread pool, schedule algs directly
• -1: max - use all available cores
• 0 or greater: requested pool size plus one
• any other negative value: error