Set the bounds in a smarter way

Right now, in an empty learner choose_points(n) will always return maximum two points (the bounds).

We need to think of a better way.

The current functionality will already break when adding a known point before the bounds are evaluated.