Don't support limitless pagination

The pagination storage function supported not specifiying a limit on the
number of events returned. This was triggered when using the search or
context API with a limit of zero, which the storage function took to
mean not being limited.

synapse/storage/stream.py

@@ -684,8 +684,7 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore):
                 results to only those before
             direction(char): Either 'b' or 'f' to indicate whether we are
                 paginating forwards or backwards from `from_key`.
-            limit (int): The maximum number of events to return. Zero or less
-                means no limit.
+            limit (int): The maximum number of events to return.
             event_filter (Filter|None): If provided filters the events to
                 those that match the filter.
 
@@ -694,6 +693,9 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore):
             as a list of _EventDictReturn and a token that points to the end
             of the result set.
         """
+
+        assert int(limit) >= 0
+
         # Tokens really represent positions between elements, but we use
         # the convention of pointing to the event before the gap. Hence
         # we have a bit of asymmetry when it comes to equalities.
@@ -723,18 +725,14 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore):
             bounds += " AND " + filter_clause
             args.extend(filter_args)
 
-        if int(limit) > 0:
         args.append(int(limit))
-            limit_str = " LIMIT ?"
-        else:
-            limit_str = ""
 
         sql = (
             "SELECT event_id, topological_ordering, stream_ordering"
             " FROM events"
             " WHERE outlier = ? AND room_id = ? AND %(bounds)s"
             " ORDER BY topological_ordering %(order)s,"
-            " stream_ordering %(order)s %(limit)s"
+            " stream_ordering %(order)s LIMIT ?"
         ) % {
             "bounds": bounds,
             "order": order,