one is unsorted and fast and we only care about it with picking a new free range.
the other is sorted and used when merging a new freed range with existing free ranges.
both are computed from the events id index at beginning, then tracked manually on each addition or deletion.
this change uncovered some errors so we fixed them and added some more fuzz test invariant checking.
code is simplified a little bit.
there was another thing I forgot.
fiatjaf