False early exits

[Zannick]

For instance, with AV2 if I provide no routes as input, the search finishes "after 775 rounds (0 dead-ends), skipped 8+1202 pushes + 0+26 pops", meaning 8 states were over the time limit (set as an unreasonable upper bound and very likely meaning the estimate did not find a valid path), and 1202+26 were duplicates.

It would be nice to have some output to debug from, such as a "best" state that was the farthest reached, or a set of spots reached, etc.

Separately, we could maybe ignore the limit if the estimate is the special value and we have no solutions yet.