RR uniqueness constraint on (rrset_id, content), version bumps, and frontend adoption updates

[peterthomassen]

About the first commit:

• During high concurrency, we sometimes end up sending PATCH requests with duplicate records to nslord, who rejects them, causing a 500 response. In all instances observed in nslord pcaps (correlated with www logs), there were to updates on the same domain within a fraction of a second.
• This happens when the second request's serializer fetches the RRset contents from the database before the first request writes them, so two rows get written.
• PDNSChangeTracker then constructs the nslord update request from the database, with duplicate records.
• This change makes this fail earlier, throwing a psycopg.errors.ExclusionViolation exception. Django turns this into an IntegrityError, which our exception handler turns into a 409 response.
• A better approach would be to lock the RRset's records for reading while a corresponding write transaction is pending. However, that will cause processes to hang, which might lead to other problems. Also, I'm not sure how to do it. I'd thus like to see if this fix suffices.

Other commits handle various routine work.

[nils-wisiol]

lgtm