enhance: better word division for highlighting

[gotoh]

Summary

Improve inline diff highlighting by replacing the ASCII delimiter–based
word division with a Unicode category–based approach, primarily to
produce more precise highlights in languages like Japanese and Chinese.

Problem

The previous implementation splits text at a fixed set of delimiter
characters (spaces, tabs, and common ASCII symbols such as +-*/=!,;).
Non-delimiter characters — including CJK ideographs, Hiragana, and
Katakana — are never treated as boundaries, so a small change within
Japanese or Chinese text causes the entire surrounding phrase to be
highlighted as changed.

Solution

Each character is classified into one of three categories. Consecutive
characters of the same category are grouped into one chunk, except for
the Other category which retains the same per-character behavior as
the previous implementation:

Category	Characters	Chunking
Letter	Latin, Greek, Cyrillic and diacritic variants (é, ü, ß…)	grouped
OtherLetter	CJK, Hiragana, Katakana, Hangul, Thai, Arabic, etc.	grouped
Other	Whitespace, punctuation, symbols — same as previous delimiters	per character

CJK punctuation (。、「」…) falls into Other and acts as a natural
boundary between OtherLetter chunks, making highlighted changes
more precise without requiring language-specific word segmentation.

Category values for all 65,536 char values are pre-computed into a
static read-only array at startup for lock-free O(1) lookup.

Japanese — before / after

Before	After

English — no regression

Before	After

Check point

• Japanese diff: divided by non CJK character mixed in sentence without space character.
• Japanese diff: CJK punctuation correctly splits chunks at sentence boundaries
• ASCII operators and punctuation: per-character precision preserved (/ vs *, == vs !=, etc.)
• European words with diacritics (café, über…): treated as single word chunks