improved get_first: check boundary before entering subtrees;

simplified join_results: no longer work "towards" deps, which simplifies task queue management and maintains strict bottom up discipline (without "transfer of priority" to required futures); more efficient Task_Queue.dequeue, with internal cache (reduces wast of cycles with many idle workers); removed complicated/expensive Task_Queue.dequeue_towards;

native get_first, with optional boundary; tuned signature;