[cdc_rsync] Improve throughput for local copies (#74)

On Windows, fclose() seems to be very expensive for large files, where
closing a 1 GB file takes up to 5 seconds. This CL calls fclose() in
background threads. This tremendously improves local syncs, e.g.
copying a 4.5 GB, 300 files data set takes only 7 seconds instead of
30 seconds.

Also increases the buffer size for copying from 16K to 128K (better
throughput for local copies), and adds a timestamp to debug and
verbose console logs (useful when comparing client and server logs).

common/threadpool_test.cc

@@ -151,5 +151,37 @@ TEST_F(ThreadpoolTest, GetCompletedTask) {
   EXPECT_EQ(completed_task.get(), task);
 }
 
+TEST_F(ThreadpoolTest, SetTaskCompletedCallback) {
+  auto task_func = [](Task::IsCancelledPredicate) { /* empty */ };
+
+  Semaphore task_finished(0);
+  Threadpool pool(1);
+  std::atomic_bool finished = false;
+  pool.SetTaskCompletedCallback(
+      [&task_finished, &finished](std::unique_ptr<Task> task) {
+        finished = true;
+        task_finished.Signal();
+      });
+  pool.QueueTask(std::make_unique<TestTask>(task_func));
+  task_finished.Wait();
+  EXPECT_TRUE(finished);
+  EXPECT_FALSE(pool.TryGetCompletedTask());
+}
+
+TEST_F(ThreadpoolTest, WaitForQueuedTasksAtMost) {
+  Semaphore task_signal(0);
+  auto task_func = [&task_signal](Task::IsCancelledPredicate) {
+    task_signal.Wait();
+  };
+  Threadpool pool(1);
+  pool.QueueTask(std::make_unique<TestTask>(task_func));
+  pool.QueueTask(std::make_unique<TestTask>(task_func));
+  EXPECT_FALSE(pool.WaitForQueuedTasksAtMost(1, absl::Milliseconds(10)));
+  task_signal.Signal();
+  EXPECT_TRUE(pool.WaitForQueuedTasksAtMost(1, absl::Milliseconds(5000)));
+  EXPECT_EQ(pool.NumQueuedTasks(), 1);
+  task_signal.Signal();
+}
+
 }  // namespace
 }  // namespace cdc_ft