It looks like the weekly docs check is failing because some external links went offline, not because of any changes I made.

Why can't we just redefine what --in-memory does?

It's a hidden field already, and I can't think of a use-case for customers, so changing the behaviour here seems fine to me (provided we note this in the release notes).

@dannykopping, many of the CLI tests currently use --in-memory (see here). We'd need to modify them, as the new approach would spawn a new Postgres process for every test, which is too heavy-weight. I was planning to make --in-memory a deprecated alias for --ephemeral once the in-memory DB is fully removed. This way, we can split the migration into smaller PRs that are easier to review.