Serve-stale "staleonly" recursion race condition

Fixes a couple of things with the serve-stale implementation.

1. Use a different variable for determining whether we can detach the client from the nmhandle. Using DNS_DBFIND_STALEONLY flag for this is information overload and makes things more confusing (first commit)
2. We can now actually simplify the logic when to detach the client (third commit)
3. Refactor the try stale lookup, make the "staleonly" scenario look more similar to the other scenarios (second commit, I apologize for the diff).
4. Add code to remove rdatasets from the message that were added on a "staleonly" lookup.
5. Remove the result == ISC_R_SUCCESS exception on "staleonly" lookups (use the same logic for "stale found" as in other stale lookup scenarios (dbfind_stale, stale_refresh_window).
6. Fix restoring the RECURSIONOK attribute after a "staleonly" lookup.
7. Bail "stale-answer-client-timeout" lookup if RPZ processing is on.

Closes #2594 (closed)