ReplicatedPG: merge apply_ctx_scrub_stats into apply_stats

ReplicatedPG: merge apply_ctx_scrub_stats into apply_stats

The only way the two aren't called essentially atomically is when the
ECBackend needs to delay a write for rmw.  In such a case, we actually
want to delay the scrub_cstat application as well since otherwise, we'd
have to flush the backend before settling up at the end of scrub.

Perhaps this should be a separate function to distinguish between the
replica+primary info update and the primary scrub_cstat update, but that
seems more complicated than it's really worth.

Fixes: http://tracker.ceph.com/issues/17999
Signed-off-by: Samuel Just <sjust@redhat.com>