I had a report from a user complaining about stale IPNS records.

After trying out a small repro on my machine, it is fairly easy to have 1h+ sync times because we will cache IPNS records for 1h without ever challenging them.

The IPNS code is surprising, it implements rollback based resolution, so it can get bad candidates early, start next recursive resolution, then when it get better candidates it can rollback and reconciliate the previous name resolution with the newer better results (or not if better candidates lines up in the existing timeline).
This allows to optimistically start with bad candidates and validate or cancel and restart with better ones later.

I think the cache should be used as a first untrusted candidate, so when a cache entry is found, we add as the first candidate in the rollback resolution process, but then we proceed with the rollback resolution as usual.

This would also combine nicely with #397 (altho any of the two is still a good improvement)