[big_whack]

I would distinguish between "local dev environment" and "debugging specific data-dependent customer issue".

My experience is you can usually write the application with no dependencies on customer data, and do local development entirely with synthetic data.

For debugging specific customer escalations or understanding data distributions for your synthetic data modeling, you can use a replica of production. No need to extract just a subset or have devs host it themselves. All access can be controlled and logged this way.

Better than a replica would be something like [1] but that isn't available everywhere.

I agree the problem of "dump a DB sample that respects FK constraints" comes up sometimes and is interesting but I'm not sure I'd use it for the problem you describe.

[1] https://devcenter.heroku.com/articles/heroku-postgres-fork