gpu: Respect sync token dependencies for UpdateSyncToken.

This was inadvertantly removed with change [1]. The reason it likely
hasn't been an issue in practice is the use-cases which rely on this
API are updates to GMBs on the CPU. We already need to wait on fences
to ensure the GPU side work is complete before writing to the memory on
the CPU so the sync token wait ends up being moot.

Still restoring the code for now since I don't understand all the
use-cases that rely on it to decide whether it can be removed.

[1]: https://chromium-review.googlesource.com/c/chromium/src/+/1611102

R=ericrk@chromium.org

Bug: 900706
Change-Id: I174716cad94bae7f33ef80d79b677af7ea1e6d6e
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1750103
Commit-Queue: Khushal <khushalsagar@chromium.org>
Commit-Queue: Eric Karl <ericrk@chromium.org>
Auto-Submit: Khushal <khushalsagar@chromium.org>
Reviewed-by: Eric Karl <ericrk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#686229}

gpu/ipc/client/shared_image_interface_proxy.cc

@@ -177,6 +177,8 @@ void SharedImageInterfaceProxy::UpdateSharedImage(
     const Mailbox& mailbox) {
   std::vector<SyncToken> dependencies;
   if (sync_token.HasData()) {
+    DCHECK(!acquire_fence);
+
     dependencies.push_back(sync_token);
     SyncToken& new_token = dependencies.back();
     if (!new_token.verified_flush()) {
@@ -202,9 +204,10 @@ void SharedImageInterfaceProxy::UpdateSharedImage(
           route_id_, mailbox, ++next_release_id_, acquire_fence_handle));
       return;
     }
-    last_flush_id_ =
-        host_->EnqueueDeferredMessage(GpuChannelMsg_UpdateSharedImage(
-            route_id_, mailbox, ++next_release_id_, acquire_fence_handle));
+    last_flush_id_ = host_->EnqueueDeferredMessage(
+        GpuChannelMsg_UpdateSharedImage(route_id_, mailbox, ++next_release_id_,
+                                        acquire_fence_handle),
+        std::move(dependencies));
   }
 }