show a hundred users' recommendations

[jhanley634]

At the April 17th 5mof we will publicly demonstrate some media recommendations. Here is a milestone along the way to that goal.

The whole thing could plausibly be completed within 7 days. Or, feel free to bite off what fits, and push subsequent steps into future sub-issues and PRs. The main branch already has a little bit of relevant supporting code.

Eventually we'll work with some number of "current" users, N, greater than 100. Start with N = 3, then 100, and we'll build upon that. Each current user reveals a handful of ratings, maybe 2 good and 1 bad. In contrast, users in the prize dataset who reveal all of their ratings to the model are "historic" users. They offer training data, but do not participate in the production of new recommendations.

1. Identify N current users, and store their revealed ratings (in a table, text file, whatever).
2. Produce half a dozen ratings for each current user, display titles on screen, and manually review / evaluate.
3. Store those ratings in a table which pipenv run browse / datasette can display. (For 100 users, that's 600 recommendation rows.)

With that in hand, we'll be in a better position to flesh out: first, a fancier "display a user's recommendations" (web?) UI, and then, a fancier data-gathering UI in which a single current user will "tell me your ratings!".