fix generation of benchmark tests from pareto grid paremeters (CAGRA)

we weren't correctly checking whether proposed tests guaranteed that intermediate-graph-degree >= graph-degree since cuvs-lucene automatically reduces intermediate-graph-degree to the value of graph-degree should intermediate-graph-degree > graph-degree. with the linked fix, we now skip proposed tests when intermediate-graph-degree > graph-degree