Optimize normalized concept generation

Brainstorming:

* Don't create new DB entries unless the new concept differs from the old one, and keep updates to a minimum
* Rather than scanning for all concept IDs, querying to build concept groups, and querying again to build the record itself, try to combine steps

No files, tests, or entry points are named. Start by locating normalized concept generation and its database writes, then compare the current queries and updates with the two optimization goals; done means avoiding unchanged concept entries and minimizing the query steps without changing the generated record.

Written by the indexing model from the issue text.