AI Controller MC Yielding (tgstation#87094)

## About The Pull Request

AI controllers do not respect their tick limits, behaving as if they
need to process all controllers in one tick.

This means every time the idle subsystem runs we process 500 controllers
at once. This is bad, and induces overtime for free, which sucks.

I can't think of a reason we would need to do all this work at once.

code/controllers/subsystem/ai_controllers.dm

@@ -8,8 +8,10 @@ SUBSYSTEM_DEF(ai_controllers)
 	runlevels = RUNLEVEL_GAME | RUNLEVEL_POSTGAME
 	///type of status we are interested in running
 	var/planning_status = AI_STATUS_ON
-	/// The tick cost of all active AI, calculated on fire.
+	/// The average tick cost of all active AI, calculated on fire.
 	var/our_cost
+	/// The tick cost of all currently processed AI, being summed together
+	var/summing_cost
 
 /datum/controller/subsystem/ai_controllers/Initialize()
 	setup_subtrees()
@@ -21,6 +23,8 @@ SUBSYSTEM_DEF(ai_controllers)
 	return ..()
 
 /datum/controller/subsystem/ai_controllers/fire(resumed)
+	if(!resumed)
+		summing_cost = 0
 	var/timer = TICK_USAGE_REAL
 	for(var/datum/ai_controller/ai_controller as anything in GLOB.ai_controllers_by_status[planning_status])
 		if(!ai_controller.able_to_plan)
@@ -30,7 +34,14 @@ SUBSYSTEM_DEF(ai_controllers)
 		if(!length(ai_controller.current_behaviors)) //Still no plan
 			ai_controller.planning_failed()
 
-	our_cost = MC_AVERAGE(our_cost, TICK_DELTA_TO_MS(TICK_USAGE_REAL - timer))
+		if(MC_TICK_CHECK)
+			break
+
+	summing_cost += TICK_DELTA_TO_MS(TICK_USAGE_REAL - timer)
+	if(MC_TICK_CHECK)
+		return
+
+	our_cost = MC_AVERAGE(our_cost, summing_cost)
 
 ///Creates all instances of ai_subtrees and assigns them to the ai_subtrees list.
 /datum/controller/subsystem/ai_controllers/proc/setup_subtrees()