Make random-query benchmark generate the same queries regardless of row and column range

[jaffee]

While trying to benchmark the affect of increasing number of rows on query time, we ran into an issue where we realized that the set of queries generated might be affected by the minimum and maximum row values since the same RNG is used both to determine the structure of the query and the arguments.

We should validate that this is a problem, and if so, fix it - probably by using a separate RNG for the row ids and the structure of the query.