Currently the release init command creates a copy of everything from all configured library source paths, into a new directory used as /repo. When we've got a lot of configured libraries, this is going to be a lot of copying.

It's unclear why this is necessary - can't we just mount the language repo as /repo (read-only)? If the idea is that the container mustn't depend on anything that's not in a source root (and not under .librarian) I'm not sure it's worth the constraint.