media/gpu/v4l2: Fix DCHECK when Initialize fails

This fixes a DCHECK that would occur if Initialize failed because then
in CreateWithOutputMode it would let the unique_ptr for the
image_processor fall out of scope and become deallocated; and we have a
sequence checker to ensure deallocation only happens on the task runner.
I would have used DeleteSoon, but that didn't seem possible because the
destructor was not public.

BUG=1050852
TEST=image_processor_test no longer DCHECKs on elm w/ 4.19 kernel

Change-Id: Ifae96c3a4206d8e94b7fdd6dd3e2ec571cc3465b
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2051162Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Commit-Queue: J Kardatzke <jkardatzke@chromium.org>
Cr-Commit-Position: refs/heads/master@{#741148}

media/gpu/v4l2/v4l2_image_processor_backend.cc

@@ -392,8 +392,15 @@ V4L2ImageProcessorBackend::CreateWithOutputMode(
                                 base::Unretained(image_processor.get()),
                                 std::move(init_cb)));
   done.Wait();
-  if (!success)
+  if (!success) {
+    // This needs to be destroyed on |backend_task_runner|.
+    backend_task_runner->PostTask(
+        FROM_HERE,
+        base::BindOnce(
+            base::DoNothing::Once<std::unique_ptr<ImageProcessorBackend>>(),
+            std::move(image_processor)));
     return nullptr;
+  }
 
   return image_processor;
 }