[ML] Persist data counts on job close before results index refresh

On job close we aim to refresh the results index before returning
so that the caller knows that all results are searchable.

It turns out that we were writing the final data counts _after_
refreshing the indices, so that a user who searched for final
data counts immediately after receiving the `_close` response
would potentially get inaccurate data counts.

This change writes the final data counts before refreshing the
indices.

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/job/process/autodetect/AutodetectCommunicator.java

@@ -170,6 +170,7 @@ public void close() {
                     killProcess(false, false);
                     stateStreamer.cancel();
                 }
+                dataCountsReporter.writeUnreportedCounts();
                 autodetectResultProcessor.awaitCompletion();
             } finally {
                 onFinishHandler.accept(null, true);
@@ -180,7 +181,6 @@ public void close() {
         try {
             future.get();
             autodetectWorkerExecutor.shutdownNow();
-            dataCountsReporter.writeUnreportedCounts();
         } catch (InterruptedException e) {
             Thread.currentThread().interrupt();
         } catch (ExecutionException e) {