Attempt to add a faster path for getting the next thread from unstaged queue

[msimberg]

Currently, if the queue of pending threads is empty on a worker thread, get_next_thread returns, the scheduler goes to add new threads from the staged to pending threads, and then calls get_next_thread again, knowing that it know (with very high probability) has a pending thread to return. This roundtrip probably adds latency. Try to make get_next_thread directly move a thread from staged to pending when it runs out of pending threads and measure the difference.