IncrementalSearch edge cases

[TomAugspurger]

A couple issues when given a small grid.

For the first, we decided that training 0 models was best, as inverse(next_time_step) was 0.

For the second, we got stuck in an infinite while True loop since

1. inverse(current_time_step) == inverse(next_time_step) was always equal (both were 0), and
2. next_time_step is increasing.

We could detect this and raise when something like start < current_time_step + 1, but ideally we would support this. I'm not sure that my fix is correct though.

[mrocklin]

Maybe the inverse function itself should never produce non-positive values? I can imagine placing the min(1, there instead, or perhaps using math.ceil rather than int

[TomAugspurger]

I think this additional if check is still necessary, even if we change inverse to be always positive. When these start off at the same value (whether it's 0 or 1), we get stuck in a cycle where next_time_step is increasing, making the inverse(current_time_step) == inverse(next_time_step) condition even more false.