We need to ensure that after combining multiplanning with CBR for planning, no regressions are visible. We'll use the synthetic benchmarks suite for that and for that we need to introduce relevant test cases for the mentioned planning combination.

At the very least we should add a scenario in which:

1. multiplanner hits EOF
2. mutiplanner fills a batch
3. multiplanner exhausts the allocated number of works and so CBR is used to plan the query.