fix: Do not starve the common ForkJoinPool with blocking listener tasks (#21342) (#21358)

Calling watchAsync without any specific executor uses the common ForkJoinPool. This is bad as all the watchers will remain in a blocking state until there is a change and will thus always use up a large part of the ForkJoinPool. If other tasks try to use the common ForkJoinPool, it might run out of concurrent threads and will no longer start any tasks at all

Co-authored-by: Artur <artur@vaadin.com>

Author: vaadin-bot

vaadin-dev-server/src/main/java/com/vaadin/base/devserver/FileWatcher.java

@@ -18,6 +18,8 @@
 import java.io.File;
 import java.io.IOException;
 import java.util.Objects;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Executors;
 
 import io.methvin.watcher.DirectoryWatcher;
 import org.slf4j.Logger;
@@ -32,6 +34,9 @@ public class FileWatcher {
 
     private DirectoryWatcher watcher;
 
+    private static ExecutorService executorService = Executors
+            .newCachedThreadPool();
+
     /**
      * Creates an instance of the file watcher for the given directory.
      * <p>
@@ -62,7 +67,7 @@ public FileWatcher(SerializableConsumer<File> onChangeConsumer,
      * Starts the file watching.
      */
     public void start() {
-        watcher.watchAsync().exceptionally((e) -> {
+        watcher.watchAsync(executorService).exceptionally((e) -> {
             getLogger().error("Error starting file watcher", e);
             return null;
         });