Known limitation: ai-blame and shallow clones

[acunniffe]

Locally git blame will show inaccurate line attribution if a related commit has not been fetched. ai-blame has the same limitation and can also show inaccurate #s if the relevant authorship logs have not been fetched. We might be able to fetch any missing authorship refs when users run ai-blame. It would be very fast and would not need to be repeated every time you run blame.