Fix race in AppOpsManager client library

If the service stays down for more than 10 seconds, client threads
will time out attempting to obtain a reference, and never release
the mutex.  This causes deadlock on the next attempt to speak
with the AppOpsManager.

Test: System without AppOpsManager service no longer hangs in camera
      code.

Change-Id: I5a8ac2f0f35345f72c3d2a45b9d1fc5a78d28a51
diff --git a/libs/binder/AppOpsManager.cpp b/libs/binder/AppOpsManager.cpp
index e4d8201..f8626cb 100644
--- a/libs/binder/AppOpsManager.cpp
+++ b/libs/binder/AppOpsManager.cpp
@@ -53,7 +53,8 @@
                 ALOGI("Waiting for app ops service");
             } else if ((uptimeMillis()-startTime) > 10000) {
                 ALOGW("Waiting too long for app ops service, giving up");
-                return NULL;
+                service = NULL;
+                break;
             }
             sleep(1);
         } else {