Preload/wrangle as much external data as possible

[jthrilly]

Following from #887

The long term solution to performance issues with loading external data is to calculate this data once at a non performance critical moment (such as starting an interview session, or when the protocol is installed), and then store the decoded and restructured data. Then, stream it in as we do with media assets.

We need to think about how this works with "true" remote assets, which are loaded by URL. #847