ITS#5129 fixes

default is 2x entry cache size

(ConfigReply) for the error messages.

_db_destroy functions.

will be gone.

Much better concurrency.

Use bi_extra for back-monitor entry points. Backends should never
directly reference each other's symbols.

leave monitor calls in place, hide contents; in the long run, enable monitor and build as static by default

  database entry in cn=monitor
- improve support for custom attribute/callback registration
  in back-monitor
- design API for registered antry/attr/callback removal in back-monitor

HEADS UP: now back-bdb requires back-monitor to be present, either static
or dynamically loaded __before__ back-bdb/back-hdb is initialized.
This works for sure if both back-bdb & back-monitor are static or dynamic
(provided back-monitor.la is loaded before back_bdb.la); it is not clear
what happens if back-bdb is static and back-monitor is dynamic.  In case,
either make back-bdb monitoring conditional, or hardwire back-monitor
much like back-config and back-ldif.

Note: this breaks LDAP_COMP_MATCH and SLAP_ZONE_MALLOC. But they
were probably broken already anyway.