Less aggressive use of Monte Carlo simulation

5000 turned out to be a too low threshold: at that complexity score traditional probability calculation is still much faster.

changelog

@@ -1,5 +1,7 @@
 Version 1.13.6+dev:
  * AI:
+   * Fix a performance regression in complex combat situations such as the
+     "Oath of Allegiance" UMC campaign.
    * Fix some Micro AIs and AI helper functions not working correctly for AI
      sides under shroud
    * New function ai_helper.find_path_with_shroud()

players_changelog

@@ -3,6 +3,10 @@ changes may be omitted). For a complete list of changes, see the main
 changelog: https://github.com/wesnoth/wesnoth/blob/master/changelog
 
 Version 1.13.6+dev:
+ * AI:
+   * Fix a performance regression in complex combat situations such as the
+     "Oath of Allegiance" UMC campaign.
+
  * Graphics:
    * Improved terrain graphics: lava.
 

src/attack_prediction.hpp

@@ -61,7 +61,7 @@ struct combatant
 #endif
 
 private:
-	static const unsigned int MONTE_CARLO_SIMULATION_THRESHOLD = 5000u;
+	static const unsigned int MONTE_CARLO_SIMULATION_THRESHOLD = 50000u;
 
 	const battle_context_unit_stats &u_;