VPP-967 ensures vlib_worker_thread_node_runtime_update() only runs once per barrier sync, just before barrier release.  This reduces the time the barrier is held, and reduces packet loss.

However, vlib_worker_thread_node_runtime_update() remains O(N), where N is the number of worker threads, because it rebuilds each thread's clone datastructures in sequence.

The proposal is to do this work in parallel, on the indiviual worker threads, just after the barrier is released