Could it be as simple as (in my case):
delete from posts inner join people on posts.author_id=people.id where people.diaspora_handle not like '%diaspora.koehn.com' and posts.created_at < '2018-10-27';
That would remove three million federated posts over a year old from my pod’s database, without touching any posts originating from users on my pod. I’m doubtful anyone is looking at content that old, so collateral damage would be minimal.