Due to compression, the different replicas of a replicated extent don't
necessarily have to take up the same amount of space - so replicated
data sector counts shouldn't be stored divided by the number of
replicas.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>