ldif: deeply reworked LDIF fields processing!

ldif: deeply reworked LDIF fields processing!

* import: now relies on item_f*t() only when it's safe
  Internal field assignation used to be very wrong what made the
  LDIF import unreliable. This is not the case anymore
* import: ability to consider additional aliases of LDIF attributes,
  eg: "zip" for "postalcode", "fax" for "facsimiletelephonenumber", ...
* import: multiple emails support (up to 2, naturally)
* import/export: added support for "description" (NOTES),
  "anniversary" and "ou" (GROUPS)
* export: replaced the attribute named "xmozillaanyphone" by
  the _more standard_ "telephonenumber"

Code has been heavily commented and hopefully opens the way
 to a better and more extensible LDIF importer.

The last chunk of that patch is a bit unclear but very simple:
  LDIF export got rid of some uneeded indirections: there's no point
  in testing LDIF_ITEM_FIELDS and doing ldif_conv_table[] translations.
  Iterating over j until ITEM_FIELDS to grab names from
  ldif_field_names[j] is enough.

Other than new features, effective changes to LDIF export
 brought by this commit are:
* attribute order is changed (eg: the URL attribute will
  displayed after MOBILEPHONE)
* no more empty ",mail:" in the "dn:" when no email exists
* export now uses the "telephonenumber" attribute
  instead of "xmozillaanyphone" previously
The LDIF import may be (undetectably) slower.

That being said, the results a the below command are now
 far more reliable.
$ diff ~/.abook/addressbook
       <( ./abook --convert --infile ~/.abook/addressbook --outformat ldif | \
          ./abook --convert --informat ldif --outformat abook )