Refactor ETL for auditability and efficiency

[adborden]

I'm looking at IL Campaign Finance's ETL as a nice example of how to structure the Makefile so it's a little more intuitive, and only rebuilds what's necesary (instead of downloading or importing everything from scratch).

DataMade also has some guidelines about how to structure ETL pipelines with some stated principles that we should follow (I think we're already most of the way there, but good to be explicit).

1. Never destroy data - treat source data as immutable, and show your work when you modify it
2. Be able to deterministically produce the final data with one command
3. Write as little custom code as possible
4. Use standard tools whenever possible
5. Keep source data under version control

I think that if we're more explicit about how we extract and transform the data, it might make it easier to spot bugs, or introduce tests/invariants along the way.

[adborden]

Another principle I would want for our site, is that the data should be easy to get at. E.g. at minimum, we should provide a link to the data source within the UI. In the ideal case, we could provide download link to the data post- tranformations and cleaning.