The trouble is that when you start inventing a distributed secret blob store you suddenly have a very different set of challenges than the ones that Nostr has already solved. Look at IPFS and Tahoe-LAFS and to some extent Scuttlebot's blobstore for some suggestions, but none of them are perfect and everyone still complains about these programs.
Really there should be a Nostr team and a team building the secret blob store, and we're risking too much technical debt being forced on the client AND relay developers.