* Split codepoint sequence normalisation out into separate function.
This creates utf8proc_normalize_utf32() which takes and returns
a UTF-32 string, applying the following options:
- UTF8PROC_NLF2LS
- UTF8PROC_NLF2PS
- UTF8PROC_NLF2LF
- UTF8PROC_STRIPCC
- UTF8PROC_COMPOSE
- UTF8PROC_STABLE
The utf8proc_reencode() function has been updated to call the
new utf8proc_normalize_utf32().
* Update code documentation: utf8proc_reencode handles UTF8PROC_CHARBOUND.
* convert sequences to utf-16 (saves 25kb)
* store sequence length in properties instead using -1 termination (saves 10kb)
* cache index for slightly faster data creation
* store lower/upper/title mapping in sequence array (saves 25kb). Add utf8proc_totitle, as title_mapping cannot be used to get the title codepoint anymore. Rename xxx_mapping to xxx_seqindex, so programs assuming a value with the old meaning fail at compile time
* change combination array data type to uint16 (saves 40kb)
* merge 1st and 2nd comb index (saves 50kb)
* kill empty prefix/suffix in combination array (saves 50kb)
* there was no need to have a separate combination start array, it can be merged in a single array
* some fixes
* mark the table as const again
* and regen
* Updates for Unicode 9.0.0 TR29 Changes
- New rules GB10/(12/13) are used to combine emoji-zwj sequences/
(force grapheme breaks every two RI codepoints). Unfortunately this
breaks statelessness of grapheme-boundary determination. Deal with
this by ignoring the problem in utf8proc_grapheme_break, and by
hacking in a special case in decompose
- ZWJ moved to its own boundclass, update what is now GB9 accordingly.
- Add comments to indicate which rule a given case implements
- The Number of bound classes Now exceeds 4 bits, expand to 8 and
reorganize fields
* Import Unicode 9 data
* Update Grapheme break API to expose state override
* Bump MAJOR version
