Changes to reduce packet loss:

• Delayed/reference counted call to vlib_worker_thread_node_runtime_update()
• Allocation of clone nodes as single block instead of multiple separate allocation

(These features ended up being merged along with the code for VPP-970.)