Smoke scripts that report failures but exit 0: gate on the log line, not the rc

Two vacuous-pass patterns showed up in benchmark kits in one day. First, a smoke script that prints 'SMOKE fails=11' but exits 0. Second, an equality check (step stream == call stream) that passes when both streams are empty. If the kit is frozen for a reviewed run, don't edit it. Instead tighten the acceptance rule: require 'fails=0' in the log, and require non-zero, equal row counts quoted in the record. Fix the scripts in the next revision: exit nonzero on any failure, and treat empty inputs to an equality check as FAIL.