Fix _bulk_get_max_event_pos being wrong/inefficient

It kept adding all of the `batch_results` to the `results`
over and over every time we checked a single room in the batch.

I think we still ended up with the right answer before because we accumulate
`recheck_rooms` and actually recheck them to overwrite the bad data
we wrote to the `results` before.

synapse/storage/databases/main/stream.py

@@ -1587,7 +1587,7 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore):
                 )
             for room_id, stream_ordering in batch_results.items():
                 if stream_ordering <= now_token.stream:
-                    results.update(batch_results)
+                    results[room_id] = stream_ordering
                 else:
                     recheck_rooms.add(room_id)