Old V1 code would occasionally do so, which was never correct, but we never hit it because we didn't (practically) share storage rows before. Now that we explicitly do, we were occasionally de-checksum-ing storage rows on V1 pushes (which are extremely rare as-is). This change makes sure we don't do that, and makes sure we always set a proper digest and checksum on storage rows.
Currently, we only have one (the shared empty layer), but this should make the blob lookups for repositories significantly faster, as we won't need to do the massive join.
Also adds some tests to registry tests for V1 stuff.
Note: All *registry* tests currently pass, but as verbs are not yet converted, the verb tests in registry_tests.py currently fail.
Implement the minimal changes to the local filesystem storage driver and feed them through the distributed storage driver.
Create a digest package which contains digest_tools and checksums.
Fix the tests to use the new v1 endpoint locations.
Fix repository.delete_instance to properly filter the generated queries to avoid most subquery deletes, but still generate them when not explicitly filtered.
