osd: automatically repair replicated replica on pulling error

osd: automatically repair replicated replica on pulling error

However this is not a very complete solution since the broken object
info may still get lost if we switch primaries or simply power off nodes.
I think a better idea would be also adding these kind of broken objects
back into replica's own missing set simultaneously, e.g., like we handling
primary reading errors.

But for now I am not sure if that should be a concern?

Fixes: http://tracker.ceph.com/issues/39101
Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>