We should add a new fetcher descendent which knows how to retry a number of times, and limited by elapsed time for the total number of retries.

This class should also know which errors exit the retry loop, using one of these strategies:

1. Retry until ErrorCodeSet contains the error
2. Retry while ErrorCodeSet contains the error
   (where the ErrorCodeSet is provided by the caller/user)

This will be needed by the new DataReplicator/InitialSync code but can also be used by sharding use of the QueryFetcher which already has its own retry loop.