Fix perf regression in PR matrix-org#3530

The get_entities_changed function was changed to return all changed
entities since the given stream position, rather than only those changed
from a given list of entities. This resulted in the function incorrectly
returning large numbers of entities that, for example, caused large
increases in database usage.

synapse/util/caches/stream_change_cache.py

@@ -74,12 +74,17 @@ def get_entities_changed(self, entities, stream_pos):
         assert type(stream_pos) is int
 
         if stream_pos >= self._earliest_known_stream_pos:
-            result = {
+            changed_entities = {
                 self._cache[k] for k in self._cache.islice(
                     start=self._cache.bisect_right(stream_pos),
                 )
             }
 
+            result = {
+                e for e in entities
+                if e in changed_entities
+            }
+
             self.metrics.inc_hits()
         else:
             result = set(entities)