[ORC] Fix synchronization in CoreAPIsTest. (#144556)

The code previously appeared to have a (benign?) race condition on
`WorkThreads.size`, since it was being accessed outside of the mutex
lock that protected it on the threads. This is usually okay since
1a1d6e6, but doesn't seem reliable in
general, so fix this code to express the intent more accurately. This
instead relies on the same general principles as ref-counting, where
each existing reference (thread) can add new references (threads)
because they already have a reference themselves (until joined).

Author: vtjnash

llvm/unittests/ExecutionEngine/Orc/CoreAPIsTest.cpp

@@ -1559,16 +1559,11 @@ TEST_F(CoreAPIsStandardTest, TestLookupWithThreadedMaterialization) {
 #if LLVM_ENABLE_THREADS
 
   std::mutex WorkThreadsMutex;
-  std::vector<std::thread> WorkThreads;
+  SmallVector<std::thread, 0> WorkThreads;
   DispatchOverride = [&](std::unique_ptr<Task> T) {
-    std::promise<void> WaitP;
     std::lock_guard<std::mutex> Lock(WorkThreadsMutex);
     WorkThreads.push_back(
-        std::thread([T = std::move(T), WaitF = WaitP.get_future()]() mutable {
-          WaitF.get();
-          T->run();
-        }));
-    WaitP.set_value();
+        std::thread([T = std::move(T)]() mutable { T->run(); }));
   };
 
   cantFail(JD.define(absoluteSymbols({{Foo, FooSym}})));
@@ -1580,8 +1575,15 @@ TEST_F(CoreAPIsStandardTest, TestLookupWithThreadedMaterialization) {
   EXPECT_EQ(FooLookupResult.getFlags(), FooSym.getFlags())
       << "lookup returned incorrect flags";
 
-  for (auto &WT : WorkThreads)
+  std::unique_lock Lock(WorkThreadsMutex);
+  // This works because every child thread that is allowed to use WorkThreads
+  // must either be in WorkThreads or its parent must be in WorkThreads.
+  while (!WorkThreads.empty()) {
+    auto WT = WorkThreads.pop_back_val();
+    Lock.unlock();
     WT.join();
+    Lock.lock();
+  }
 #endif
 }