Allow MigrateExecutable to be constructed without a message object

Apologies for sticking my nose in if this isn't the kinda solution you're looking for, but this will at least get the tests passing.

I'm a little uncomfortable with mixing getMock() and prophesize(), but that's what I went with.

Let's add a comment why we are passing in the event_dispatcher, or at least add a test of it in use. I think I know why, but it isn't blatantly obvious either.

I don't know what I was thinking in #5. RTBC. MigrateMessage is no longer required. And it is BC.

1. +++ b/core/modules/migrate/src/MigrateExecutable.php
   @@ -96,17 +96,17 @@ class MigrateExecutable implements MigrateExecutableInterface {
   +    $this->eventDispatcher = $event_dispatcher ?: \Drupal::service('event_dispatcher');
   

   This code makes some of

     /**
      * Gets the event dispatcher.
      *
      * @return \Symfony\Component\EventDispatcher\EventDispatcherInterface
      */
     protected function getEventDispatcher() {
       if (!$this->eventDispatcher) {
         $this->eventDispatcher = \Drupal::service('event_dispatcher');
       }
       return $this->eventDispatcher;
     }
   

   Unnecessary. But I don't think that making this change is necessary here. Why is it being done?

2. +++ b/core/modules/migrate/tests/src/Unit/MigrateExecutableMemoryExceededTest.php
   @@ -53,8 +54,9 @@ protected function setUp() {
   +    $this->executable = new TestMigrateExecutable($this->migration, $this->message, $event_dispatcher);
   
   +++ b/core/modules/migrate/tests/src/Unit/process/UrlEncodeTest.php
   @@ -56,7 +56,8 @@ public function testUrls($input, $output) {
   +    $executable = new MigrateExecutable($this->getMigration(), NULL, $event_dispatcher);
   

   Why is it necessary to inject the event dispatcher here suddenly? It's not obvious at all to me.

Patch in #9 no longer applies. Re-rolled.

Avoid calling setMessage() with a NULL parameter.

This looks good again.

Straightforward DX improvement. The number of places we're no longer constructing an empty message object shows that this cleanup is worthwhile, and it's totally BC. Nice work!

Committed and pushed to 8.5.x. I didn't backport it because it's an API cleanup. I also published the CR (after retitling it a bit and bumping the branch to 8.5.x).

Reverted and recommitted to fix (a) the branch and (b) a crediting issue.