The current inter-shard move is implemented as a `git fetch` from the
original shard to an empty repository on the new shard. The repository
on the new shard lacks bitmaps, and may be less well packed, than the
repository on the old shard. This has a measurable performance impact.

We may be changing from "FetchInternalRemote" to "ReplicateRepository"
in the future, but until then, forcing housekeeping to run is a good
performance optimisation - we trade off some short-term I/O load on the
new shard for better performance across a wide range of RPCs, including
an order-of-magnitude improvement in `IsAncestor`, which is the find
that motivated this change.