SIL Optimizer: add the OptimizeHopToExecutor in the late pipeline.

This is important to remove redundant `hop_to_executor` instructions after inlining.

rdar://95796233

Author: eeckstein

lib/SILOptimizer/PassManager/PassPipeline.cpp

@@ -793,6 +793,7 @@ static void addLateLoopOptPassPipeline(SILPassPipelinePlan &P) {
 // - don't require IRGen information.
 static void addLastChanceOptPassPipeline(SILPassPipelinePlan &P) {
   // Optimize access markers for improved IRGen after all other optimizations.
+  P.addOptimizeHopToExecutor();
   P.addAccessEnforcementReleaseSinking();
   P.addAccessEnforcementOpts();
   P.addAccessEnforcementWMO();

test/Concurrency/Runtime/exclusivity_custom_executors.swift

@@ -376,6 +376,7 @@ struct Runner {
 
                 await withExclusiveAccessAsync(to: &global) {
                     @MyMainActorWithAccessInUnownedExecAccessor (x: inout Int) async -> Void in
+                    print("do something to avoid optimizing away to executor switch")
                     debugLog("==> Making sure can push/pop access")
                 }
                 // In order to test that we properly hand off the access, we