Minor perf speed up for large accounts on SSS (#17751)

This works as instead of passing *all* rooms to `record_sent_rooms` we
only need to pass rooms that were previously not in the LIVE state.

This came from a py-spy where we were spending ~10% CPU calling these
functions. Note that `record_sent_rooms` is a no-op for rooms that are
already in the `LIVE` state, so we only need to call them for
`PREVIOUSLY` or `INITIAL` rooms.

changelog.d/17751.misc

@@ -0,0 +1 @@
+Minor performance increase for large accounts using sliding sync.

synapse/handlers/sliding_sync/extensions.py

@@ -572,7 +572,8 @@ async def handle_previously(room_id: str) -> None:
 
             # Now record which rooms are now up to data, and which rooms have
             # pending updates to send.
-            new_connection_state.account_data.record_sent_rooms(relevant_room_ids)
+            new_connection_state.account_data.record_sent_rooms(previously_rooms.keys())
+            new_connection_state.account_data.record_sent_rooms(initial_rooms)
             missing_updates = (
                 all_updates_since_the_from_token.keys() - relevant_room_ids
             )
@@ -763,9 +764,10 @@ async def handle_previously_room(room_id: str) -> None:
 
                 room_id_to_receipt_map[room_id] = {"type": type, "content": content}
 
-        # Now we update the per-connection state to track which receipts we have
-        # and haven't sent down.
-        new_connection_state.receipts.record_sent_rooms(relevant_room_ids)
+            # Update the per-connection state to track which rooms we have sent
+            # all the receipts for.
+            new_connection_state.receipts.record_sent_rooms(previously_rooms.keys())
+            new_connection_state.receipts.record_sent_rooms(initial_rooms)
 
         if from_token:
             # Now find the set of rooms that may have receipts that we're not sending