(Re-)Compute data checksums before insertion

Up until now, in git repositories, we have trusted the external "source" for object identifiers. Other kinds of repositories would compute the checksum "locally" before sending the data.

Before importing git repositories, we ran git-fsck on the repositories to make sure the checksums were valid.

As we're moving towards a more distributed development model, and to reduce the need for swh/devel/experiments/swh-db-audit#75 (moved) in the future, we should make swh.storage recompute identifiers, and reject the insertion of data where they don't match.

---

Migrated from T399 (view on Phabricator)