Deborgified the app cache.

Improved Andrew's hack to create temporary app caches to handle
migrations. Now the main app cache has a "master" flag set to True
(which is a non-default keyword argument, thus unlikely to be used by
mistake). Other app cache instances have "master" set to False.

The only sanctioned way to access the app cache is by importing
django.core.apps.app_cache.

If you were instanciating an app cache and relying on the Borg pattern,
you'll have to refactor your code.

tests/app_cache/tests.py

@@ -1,7 +1,7 @@
 from __future__ import absolute_import
 
 from django.core.apps import app_cache
-from django.core.apps.cache import BaseAppCache
+from django.core.apps.cache import AppCache
 from django.db import models
 from django.test import TestCase
 
@@ -30,7 +30,7 @@ class AppCacheTests(TestCase):
         old_models = app_cache.get_models(app_cache.get_app_config("app_cache").models_module)
         # Construct a new model in a new app cache
         body = {}
-        new_app_cache = BaseAppCache()
+        new_app_cache = AppCache()
         meta_contents = {
             'app_label': "app_cache",
             'app_cache': new_app_cache,
@@ -45,3 +45,10 @@ class AppCacheTests(TestCase):
             app_cache.get_models(app_cache.get_app_config("app_cache").models_module),
         )
         self.assertEqual(new_app_cache.get_model("app_cache", "SouthPonies"), temp_model)
+
+    def test_singleton_master(self):
+        """
+        Ensures that only one master app cache can exist.
+        """
+        with self.assertRaises(RuntimeError):
+            AppCache(master=True)