fix(sessds): use milder log level for regular error conditions

Especially when such events are emitted in (potentially) tight loops,
i.e. pulling messages from iterators pointing to remote shards.

apps/emqx/src/emqx_persistent_session_ds.erl

@@ -622,7 +622,7 @@ replay_streams(Session0 = #{replay := [{StreamKey, Srs0} | Rest]}, ClientInfo) -
             replay_streams(Session#{replay := Rest}, ClientInfo);
         {error, recoverable, Reason} ->
             RetryTimeout = ?TIMEOUT_RETRY_REPLAY,
-            ?SLOG(warning, #{
+            ?SLOG(debug, #{
                 msg => "failed_to_fetch_replay_batch",
                 stream => StreamKey,
                 reason => Reason,
@@ -925,7 +925,7 @@ new_batch({StreamKey, Srs0}, BatchSize, Session0 = #{s := S0}, ClientInfo) ->
             Session#{s => S};
         {error, Class, Reason} ->
             %% TODO: Handle unrecoverable error.
-            ?SLOG(info, #{
+            ?SLOG(debug, #{
                 msg => "failed_to_fetch_batch",
                 stream => StreamKey,
                 reason => Reason,

apps/emqx/src/emqx_persistent_session_ds_stream_scheduler.erl

@@ -220,7 +220,7 @@ ensure_iterator(TopicFilter, StartTime, SubId, SStateId, {{RankX, RankY}, Stream
                     },
                     emqx_persistent_session_ds_state:put_stream(Key, NewStreamState, S);
                 {error, recoverable, Reason} ->
-                    ?SLOG(warning, #{
+                    ?SLOG(debug, #{
                         msg => "failed_to_initialize_stream_iterator",
                         stream => Stream,
                         class => recoverable,