[cluster] keep track of node counts cluster-wide.

This generalizes exchange of signals between the ranks using a non-blocking all-reduce. It is now used for the stop signal and the node count, but should be easily generalizable (TB hits, and ponder still missing). It avoids having long-lived outstanding non-blocking collectives (removes an early posted Ibarrier). A bit too short a test, but not worse than before:

Score of new-r4-1t vs old-r4-1t: 459 - 401 - 1505  [0.512] 2365
Elo difference: 8.52 +/- 8.43

src/thread.cpp

@@ -163,7 +163,6 @@ void ThreadPool::start_thinking(Position& pos, StateListPtr& states,
   main()->wait_for_search_finished();
 
   stopOnPonderhit = stop = false;
-  Cluster::sync_start();
 
   ponder = ponderMode;
   Search::Limits = limits;
@@ -201,5 +200,7 @@ void ThreadPool::start_thinking(Position& pos, StateListPtr& states,
 
   setupStates->back() = tmp;
 
+  Cluster::signals_init();
+
   main()->start_searching();
 }