Is trace really an iterator?

[joshrule]

We currently use Iterator to extend a Trace out from a root. The similarity between the previous step-based implementation and Iterator's next made it an easy implementation. I'm increasingly convinced, however, that we either need to remove the Iterator implementation or rework the way in which it is constructed. The overall semantics of Trace is confusing and makes it possible to compute values that the user might not intend. For example, you can build a trace with 100 nodes and then ask for the likelihood of a given rewrite in a 50 node trace, but because 100 nodes already exist, the likelihood will be over all 100 nodes rather than just 50.