Ash should automatically define join resources for many-to-many relationships.

This one is a bit weird, because if the data layer isn't postgres do we even _need_ a join resource? if both sides are ETS backed (for example) then can we just store arrays or something?

The issue names no files, tests, or entry points. Start by clarifying how many-to-many relationships should behave across PostgreSQL and ETS-backed data layers, including whether a join resource is always required. Done should be a documented, agreed design with explicit behavior for both cases.

Written by the indexing model from the issue text.