Always cache 'event_to_prev_state_group' (#9950)

Fixes regression in send PDU times introduced in #9905.

Author: erikjohnston

changelog.d/9950.feature

@@ -0,0 +1 @@
+Improve performance of sending events for worker-based deployments using Redis.

synapse/handlers/message.py

@@ -1050,19 +1050,20 @@ async def cache_joined_hosts_for_event(
         )
 
         if state_entry.state_group:
+            await self._external_cache.set(
+                "event_to_prev_state_group",
+                event.event_id,
+                state_entry.state_group,
+                expiry_ms=60 * 60 * 1000,
+            )
+
             if state_entry.state_group in self._external_cache_joined_hosts_updates:
                 return
 
             joined_hosts = await self.store.get_joined_hosts(event.room_id, state_entry)
 
             # Note that the expiry times must be larger than the expiry time in
             # _external_cache_joined_hosts_updates.
-            await self._external_cache.set(
-                "event_to_prev_state_group",
-                event.event_id,
-                state_entry.state_group,
-                expiry_ms=60 * 60 * 1000,
-            )
             await self._external_cache.set(
                 "get_joined_hosts",
                 str(state_entry.state_group),