Avoid retrying the same solutions

Partially fixes #357.
Instead of starting a new epoch immediately at the end of the last
one, start epochs when new solution sets are loaded. Scenario
index is set at instantiation and does not reset with new epoch.

mpisppy/cylinders/xhatshufflelooper_bounder.py

@@ -139,6 +139,8 @@ def _vb(msg):
                 # update the caches
                 self.opt._put_nonant_cache(self.localnonants)
                 self.opt._restore_nonants()
+
+                scenario_cycler.begin_epoch()
 
             next_scendict = scenario_cycler.get_next()
             if next_scendict is not None:
@@ -147,8 +149,6 @@ def _vb(msg):
                 if update:
                     _vb(f"   Updating best to {next_scendict}")
                     scenario_cycler.best = next_scendict
-            else:
-                scenario_cycler.begin_epoch()
 
             #_vb(f"    scenario_cycler._scenarios_this_epoch {scenario_cycler._scenarios_this_epoch}")
 
@@ -174,6 +174,7 @@ def __init__(self, shuffled_scenarios,nonleaves,reverse,iter_step):
         self._shuffled_scenarios = shuffled_scenarios
         self._num_scenarios = len(shuffled_scenarios)
 
+        self._cycle_idx = 0
         self._begin_normal_epoch()
 
         self._best = None
@@ -249,7 +250,6 @@ def _begin_normal_epoch(self):
             self._reversed = False
         self._shuffled_snames = [s[1] for s in self._shuffled_scenarios]
         self._original_order = [s[0] for s in self._shuffled_scenarios]
-        self._cycle_idx = 0
         self._cur_ROOTscen = self._shuffled_snames[0]
         self.create_nodescen_dict()
 
@@ -259,7 +259,6 @@ def _begin_reverse_epoch(self):
         self._reversed = True
         self._shuffled_snames = [s[1] for s in reversed(self._shuffled_scenarios)]
         self._original_order = [s[0] for s in reversed(self._shuffled_scenarios)]
-        self._cycle_idx = 0
         self._cur_ROOTscen = self._shuffled_snames[0]
         self.create_nodescen_dict()