Update to 0.9.4.

New in 0.9.4:
* The data tables and line breaking algorithm have been updated to Unicode
  version 6.0.0.
* A new include file unigbrk.h is provided. It declares functions for
  grapheme cluster breaking, that is, determining the boundaries between
  graphemes. See the documentation chapter "Grapheme cluster breaks in strings"
  for details.
* In the include file unictype.h, constants are defined for the group of
  general categories LC ("Cased Letter").
* In the include file unictype.h, functions for associating canonical
  combining classes with names have been added:
    uc_combining_class_name
    uc_combining_class_long_name
    uc_combining_class_byname
* In the include file unictype.h, functions for the Arabic joining type and
  the Arabic joining group have been added:
    uc_joining_type_name
    uc_joining_type_long_name
    uc_joining_type_byname
    uc_joining_type
    uc_joining_group_name
    uc_joining_group_byname
    uc_joining_group
* In the include file unictype.h, functions for new predefined properties
  have been added:
    uc_is_property_cased
    uc_is_property_case_ignorable
    uc_is_property_changes_when_lowercased
    uc_is_property_changes_when_uppercased
    uc_is_property_changes_when_titlecased
    uc_is_property_changes_when_casefolded
    uc_is_property_changes_when_casemapped
  But it's recommended to use the case mapping functions from unicase.h
  instead.
* In the include file unictype.h, the functions for bidi class, formerly known
  as bidirectional category, have been renamed:
    uc_bidi_category_name   -> uc_bidi_class_name
    uc_bidi_category_byname -> uc_bidi_class_byname
    uc_bidi_category        -> uc_bidi_class
    uc_is_bidi_category     -> uc_is_bidi_class
  The old function names still exist, but are obsolete.
* In the include file unictype.h, functions for returning long names of
  property values have been added:
    uc_general_category_long_name
    uc_bidi_class_long_name
  The functions
    uc_general_category_byname
    uc_bidi_class_byname
  have been extended to recognize long names as well as short names.
* It is now easier to detect the subminor version: The value of the variable
  _libunistring_version and of the macro _LIBUNISTRING_VERSION now includes
  also the subminor version.
* The functions u8_mbtouc and u8_mbtouc_unsafe now handle ill-formed UTF-8
  input in a better way, that is more compliant with W3C recommendations.
* The functions u8_strcoll, u16_strcoll, u32_strcoll now produce results that
  are less dependent on the iconv implementation in use.
* The functions u8_strstr, u16_strstr, u32_strstr now perform in O(n) time
  worst-case, where n is the sum of the lengths of the argument strings.


(wiz)