Fabric: Ressetting callable inside `JMessageQueueThread`

Summary:
We recently realized that `JNativeRunnable` instances that RN uses to pass C++ callables to Java land actually are GC managed objects. That makes their lifetime quite unpredictable (longer than necessary). Normally, it's fine but some C++ code explicitly relies on deallocation order.
To make the behavior of `JMessageQueueThread` more predictable, now we clear/reset stored `std::function` object right after an invocation to explicitly free all associated resources (`JNativeRunnable` still holds some wrapper but that wrapper holds nothing).

Changelog: [INTERNAL]

Reviewed By: JoshuaGross

Differential Revision: D18603390

fbshipit-source-id: 362f6cc0901cbe14d3360b928c98e204d277b1aa

ReactAndroid/src/main/jni/react/jni/JMessageQueueThread.cpp

@@ -32,9 +32,20 @@ struct JavaJSException : jni::JavaClass<JavaJSException, JThrowable> {
 };
 
 std::function<void()> wrapRunnable(std::function<void()>&& runnable) {
-  return [runnable=std::move(runnable)] {
+  return [runnable = std::move(runnable)]() mutable {
+    if (!runnable) {
+      // Runnable is empty, nothing to run.
+      return;
+    }
+
+    auto localRunnable = std::move(runnable);
+
+    // Clearing `runnable` to free all associated resources that stored lambda
+    // might retain.
+    runnable = nullptr;
+
     try {
-      runnable();
+      localRunnable();
     } catch (const jsi::JSError& ex) {
       throwNewJavaException(
           JavaJSException::create(ex.getMessage().c_str(), ex.getStack().c_str(), ex)