I've run into a problem while hacking operational attributes collection
in slap_send_search_entry() when the "rwm" overlay is stacked.
For bdb/hdb, the be_operational() handler requires the entry to have
an appropriate e_private field, but rwm makes a copy. I could
add the original e_private value to the copy, for that purpose,
and clean it up before entry_free(), but unfortunately entry_free(),
if required, is called before the stack of cleanup callbacks is
invoked. If there's no drawbacks, I'd move cleanup callbacks invocation
before conditional SlapReply cleanup, so that specialized cleanup
can occur before the default catchall is run. This seems to be
more logical as well.