* feat: improve search tokenization for CJK languages
Enhance the encoder function to properly tokenize CJK (Chinese, Japanese,
Korean) characters while maintaining English word tokenization. This fixes
search issues where CJK text was not searchable due to whitespace-only
splitting.
Changes:
- Tokenize CJK characters (Hiragana, Katakana, Kanji, Hangul) individually
- Preserve whitespace-based tokenization for non-CJK text
- Support mixed CJK/English content in search queries
This addresses the CJK search issues reported in #2109 where Japanese text
like "て以来" was not searchable because the encoder only split on whitespace.
Tested with Japanese, Chinese, and Korean content to verify character-level
tokenization works correctly while maintaining English search functionality.
* perf: optimize CJK search encoder with manual buffer tracking
Replace regex-based tokenization with index-based buffer management.
This improves performance by ~2.93x according to benchmark results.
- Use explicit buffer start/end indices instead of string concatenation
- Replace split(/\s+/) with direct whitespace code point checks
- Remove redundant filter() operations
- Add CJK Extension A support (U+20000-U+2A6DF)
Performance: ~878ms → ~300ms (100 iterations, mixed CJK/English text)
* test: add comprehensive unit tests for CJK search encoder
Add 21 unit tests covering:
- English word tokenization
- CJK character-level tokenization (Japanese, Korean, Chinese)
- Mixed CJK/English content
- Edge cases
All tests pass, confirming the encoder correctly handles CJK text.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
ec26ebcc9e