Fix atomic access to deduplication stats

Also some imprecissions coming from error cases are handled
now properly.

Signed-off-by: Jacobo Giralt <jacobo.giralt@bsc.es>