Reasoner: adding limit should lead to flat, not linear, performance

[flyingsilverfin]

Adding a limit to the various actions to do with reasoning, like limit 3 should squash the linear increases in traversal time as the simulation grows the data set, especially once we remove the negations from the simulation agents.

Notably, we still see a linear increase in the reasoning-related queries, with a limit and without negations. We should investigate the cause of this for production.