Refactor fetcher/parser classes to make fetcher the iterator

My vision was to make the fetcher a format-agnostic iterator that produces a chunk of raw data corresponding to one migratable data item at a time, which the parser would then parse into individual fields. Once I express that clearly this way, however, a problem becomes immediately apparent - for most data formats (like, say, XML or JSON) you have to know how to parse the format to identify those chunks to be further parsed, making the fetcher no longer format-agnostic. We might have a mechanism for the format-agnostic fetcher to call the format-savvy parser to do the work, but this ties the fetcher and parser more closely together. At any rate, for the XMLReader source which is our current practical use case, it is most convenient for it to do both kinds of parsing as it goes - separating the migratable items, and parsing out the specific fields from each item.

So, at this point, after periodically puzzling over how to cleanly support both grab-everything-and-give-it-to-the-parser and take-a-piece-at-a-time, I'm ready to move on. In particular, the XMLReader approach inherently mixes parsing and fetching so they're difficult to cleanly separate. So, I'm postponing this rather than closing it in case someone has a good idea to generalize it, but for I think we should stick with what we have now, where you can have a parser that doesn't need an external fetcher (i.e., XMLReader).