Use a star schema model for the datasette logs

[bendnorman]

Right now the datasette logs produced by the ELT pipeline live in a flat denormalized table. This result in some duplicate information and could result in a very wide table if we continue to add columns.

We could use star schemas to model the datasette logs. Each transaction is a log with an http request, timestamp and size. We could create a dimension table for the ip address information. This is where most of the duplicate information is because ip addresses make multiple requests.