This will eliminate the cumbersome need to use a thread just to schedule a ShardRegistry lookup (when incorporating RSM changes).

ie. instead of returning _cache->acquire(...), _getData() can return _cache->acquireAsync(...), and no longer need to take an opCtx (the VectorClock can be obtained via _service instead). Most callers will then just call .get(opCtx) on the result from _getData(). But callers such as the above, which are just scheduling a lookup and aren't interested in the result, don't need to do that.