Devise more versatile trust tree

Currently `Recipe`s define trust via a simple whitelist. The issue here is that trust is not binary, and recipe would be better off specifying the capacities in which they trust each author. E.g. "I trust author X with documents named `championship/some_fixed_id/match/*`", or "I trust author Y with external ID attestations referencing platform A".

It might also be useful to move the trust list on to a separate entity, making it possible to execute the same recipe against different authors in order to compare the outcome.