UpdateAttempterAndroid::Init initiates merge

On update_engine starts, schedule CleanupPreviousUpdateAction that calls
CleanupSuccessfulUpdate to do necessary cleanup as soon as possible.

In the good case, update_engine initiates merge when
sys.boot_completed, and clean up snapshots.

If the update is
rolled back or partitions are flashed, the following happens (on
a Virtual A/B device):

- UpdateAttempterAndroid::CleanupSuccessfulUpdate is called
  - DynamicPartitionControlAndroid::CleanupSuccessfulUpdate is called
  - SnapshotManager::InitiateMergeAndWait is called
  - SnapshotManager::RemoveAllUpdateState(before_cancel) is called
  - before_cancel is called,
    DeltaPerformer::ResetUpdateProgress is called
    - All update states in update_engine is reset.
  - SnapshotManager proceeds to delete snapshots
    - All update states in SnapshotManager is reset.

Hence, on an VAB device, when an update is rolled back or partitions
are flashed, the whole update needs to be re-applied
(while in A/B, it skips writing and directly start verifying hashes of
the target partitions because the update markers are still there).

Bug: 147696014
Test: apply OTA then reboot, inspect logs and do `snapshotctl dump`

Change-Id: I0fc5e7768dfb53e4fd474f2d8d85d2a1b615a88b
diff --git a/dynamic_partition_control_android.h b/dynamic_partition_control_android.h
index e7ae26b..aff0296 100644
--- a/dynamic_partition_control_android.h
+++ b/dynamic_partition_control_android.h
@@ -46,6 +46,10 @@
                                   uint64_t* required_size) override;
   bool FinishUpdate() override;
   ErrorCode CleanupSuccessfulUpdate() override;
+  std::unique_ptr<AbstractAction> GetCleanupPreviousUpdateAction(
+      BootControlInterface* boot_control,
+      PrefsInterface* prefs,
+      CleanupPreviousUpdateActionDelegateInterface* delegate) override;
 
   // Return the device for partition |partition_name| at slot |slot|.
   // |current_slot| should be set to the current active slot.