In Spark's Structured Streaming, what is the default output mode for a query targeting a fault-tolerant table where existing rows remain unchanged?