avoid update_workers_node calls per request if using CacheShareFor

If using the newer CacheShareFor setting, we want to use cached and possibly stale states for a time to optimize performance. But update_workers_node calls could still be repeated per request in proxy_cluster_trans and internal_find_best_by_requests.

Perhaps we can similarly leave any update_workers_node calls to only occur later in the watchdog thread if using CacheShareFor to further optimize performance.