Disable partial state group caching for wildcard lookups

When _get_state_for_groups is given a wildcard filter, just do a complete
lookup. Hopefully this will give us the best of both worlds by not filling up
the ram if we only need one or two keys, but also making the cache still work
for the federation reader usecase.

synapse/util/caches/dictionary_cache.py

@@ -107,29 +107,28 @@ class DictionaryCache(object):
         self.sequence += 1
         self.cache.clear()
 
-    def update(self, sequence, key, value, full=False, known_absent=None):
+    def update(self, sequence, key, value, fetched_keys=None):
         """Updates the entry in the cache
 
         Args:
             sequence
-            key
-            value (dict): The value to update the cache with.
-            full (bool): Whether the given value is the full dict, or just a
-                partial subset there of. If not full then any existing entries
-                for the key will be updated.
-            known_absent (set): Set of keys that we know don't exist in the full
-                dict.
+            key (K)
+            value (dict[X,Y]): The value to update the cache with.
+            fetched_keys (None|set[X]): All of the dictionary keys which were
+                fetched from the database.
+
+                If None, this is the complete value for key K. Otherwise, it
+                is used to infer a list of keys which we know don't exist in
+                the full dict.
         """
         self.check_thread()
         if self.sequence == sequence:
             # Only update the cache if the caches sequence number matches the
             # number that the cache had before the SELECT was started (SYN-369)
-            if known_absent is None:
-                known_absent = set()
-            if full:
-                self._insert(key, value, known_absent)
+            if fetched_keys is None:
+                self._insert(key, value, set())
             else:
-                self._update_or_insert(key, value, known_absent)
+                self._update_or_insert(key, value, fetched_keys)
 
     def _update_or_insert(self, key, value, known_absent):
         # We pop and reinsert as we need to tell the cache the size may have