use liblutil's ldif_read_record instead of read_one_record
  track line numbers across entire LDIF instead of just for a single
	entry, to allow easier locating of errors. (we can also add the
	-j jump option later.)
  split Add/Modify parse paths, precompute size of pmods instead of
  	continuously using realloc.

Entry processing is now linear time instead of n^2 with entry size.