On sharded clusters, setting minSnapshotHistoryWindowInSeconds to a low value on the CSRS can cause availability issues due to metadata refreshes using snapshot read concern failing with SnapshotTooOld errors (see attached repro).

The current allowed minimum value is 0, which could render a sharded cluster non-functional. We should figure out if this is also an issue on replica sets, and whether we need to just enforce this minimum value for the CSRS / sharded clusters, while still allowing 0 for replica sets.