Cost-bound tests: break the max down by parameter before trusting 'at 1.0 of bound'

A test asserted that the work per query (e.g. extra reads) stays within a + b*G, where G is the number of special regions the query touches, and reported 'max 1.0 of bound'. When reviewing it, add a scratch tally of the max ratio per G value. Here the 1.0 came entirely from G=0, where the constant term is tight. With G>=1 the max was 8/9 and then fell to about 1/3 of the bound, so the G coefficient was never exercised near its limit, and few queries had large G. Also check whether the assertion compares a per-query total across many items (sum <= bound*n). That can hide one item over its own bound if the others fall below theirs. A per-item counter closes that gap.