[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/cluster.h

@@ -74,9 +74,10 @@ int rank();
 inline bool is_root() { return rank() == 0; }
 void save(Thread* thread, TTEntry* tte, Key k, Value v, Bound b, Depth d, Move m, Value ev);
 void pick_moves(MoveInfo& mi);
-void sum(uint64_t& val);
-void sync_start();
-void sync_stop();
+uint64_t nodes_searched();
+void signals_init();
+void signals_poll();
+void signals_sync();
 
 #else
 
@@ -88,9 +89,10 @@ constexpr int rank() { return 0; }
 constexpr bool is_root() { return true; }
 inline void save(Thread*, TTEntry* tte, Key k, Value v, Bound b, Depth d, Move m, Value ev) { tte->save(k, v, b, d, m, ev); }
 inline void pick_moves(MoveInfo&) { }
-inline void sum(uint64_t& ) { }
-inline void sync_start() { }
-inline void sync_stop() { }
+uint64_t nodes_searched();
+inline void signals_init() { }
+inline void signals_poll() { }
+inline void signals_sync() { }
 
 #endif /* USE_MPI */